Creating a Multi-AI Deploy App with Multiple Models¶
This tutorial shows how to create an inference application with multiple models, focusing on model files organization, accessing and inferring with named model network in the application, and finally building an app package.
Typically multiple models will work in tandem, e.g. a lung segmentation model’s output, along with the original image, are then used by a lung nodule detection and classification model. There is, however, a lack of such models in the MONAI Model Zoo as of now. So, for illustration purpose, two independent models will be used in this example, Spleen Segmentation and Pancreas Segmentation, both are trained with DICOM images of CT modality, and both are packaged in the MONAI Bundle format. A single input of a CT Abdomen DICOM Series can be used for both models within the application.
Important Steps¶
Place the model TorchScripts in a defined folder structure, see below for details
Pass the model name to the inference operator instance in the app
Connect the input to and output from the inference operators, as required by the app
Required Model File Organization¶
The model files in TorchScript, be it MONAI Bundle compliant or not, must each be placed in an uniquely named folder. The name of this folder becomes the name of the loaded model network in the application, and is used by the application to retieve the network via the execution context.
The folders containing the individual model file must then be placed under a parent folder. The name of this folder is chosen by the application developer.
The path of the aforementioned parent folder is used to set the well-known environment variable for the model path,
HOLOSCAN_MODEL_PATH
, when the application is directly run as a program.When the application is packaged as an MONAI Application Package (MAP), the parent folder is used as the model path, and the Packager copies all of the sub folders to the well-known
models
folder in the MAP.
Example Model File Organization¶
In this example, the models are organized as shown below.
multi_models
├── pancreas_ct_dints
│ └── model.ts
└── spleen_ct
└── model.ts
Please note,
The
multi_models
is the parent folder, whose path is used to set the well-known environment variable for model path. When using App SDK CLI Packager to build the application package, this is the used as the path for models.The sub-folder names become model network names,
pancreas_ct_dints
andspleen_model
, respectively.
In the following sections, we will demonstrate how to create and package the application using these two models.
Note
The two models are both MONAI bundles, published in MONAI Model Zoo
The DICOM CT series used as test input is downloaded from TCIA, CT Abdomen Collection ID CPTAC-PDA
Subject ID C3N-00198
.
Both the DICOM files and the models have been packaged and shared on Google Drive.
Creating Operators and connecting them in Application class¶
We will implement an application that consists of seven Operators:
DICOMDataLoaderOperator:
Input(dicom_files): a folder path (
Path
)Output(dicom_study_list): a list of DICOM studies in memory (List[
DICOMStudy
])
DICOMSeriesSelectorOperator:
Input(dicom_study_list): a list of DICOM studies in memory (List[
DICOMStudy
])Input(selection_rules): a selection rule (Dict)
Output(study_selected_series_list): a DICOM series object in memory (
StudySelectedSeries
)
DICOMSeriesToVolumeOperator:
Input(study_selected_series_list): a DICOM series object in memory (
StudySelectedSeries
)Output(image): an image object in memory (
Image
)
MonaiBundleInferenceOperator x 2:
DICOMSegmentationWriterOperator x2:
Input(seg_image): a segmentation image object in memory (
Image
)Input(study_selected_series_list): a DICOM series object in memory (
StudySelectedSeries
)Output(dicom_seg_instance): a file path (
Path
)
Note
The DICOMSegmentationWriterOperator
needs both the segmentation image as well as the original DICOM series for reusing the patient demographics and other DICOM Study level attributes, as well as referring to the original SOP instance UID.
The workflow of the application is illustrated below.
Setup environment¶
# Install MONAI and other necessary image processing packages for the application
!python -c "import monai" || pip install --upgrade -q "monai"
!python -c "import torch" || pip install -q "torch>=1.12.0"
!python -c "import numpy" || pip install -q "numpy>=1.21"
!python -c "import nibabel" || pip install -q "nibabel>=3.2.1"
!python -c "import pydicom" || pip install -q "pydicom>=1.4.2"
!python -c "import highdicom" || pip install -q "highdicom>=0.18.2"
!python -c "import SimpleITK" || pip install -q "SimpleITK>=2.0.0"
# Install MONAI Deploy App SDK package
!python -c "import monai.deploy" || pip install -q "monai-deploy-app-sdk"
Note: you may need to restart the Jupyter kernel to use the updated packages.
Download/Extract input and model/bundle files from Google Drive¶
# Download the test data and MONAI bundle zip file
!pip install gdown
!gdown "https://drive.google.com/uc?id=1llJ4NGNTjY187RLX4MtlmHYhfGxBNWmd"
# After downloading ai_spleen_bundle_data zip file from the web browser or using gdown,
!unzip -o "ai_multi_model_bundle_data.zip"
Requirement already satisfied: gdown in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (5.1.0)
Requirement already satisfied: beautifulsoup4 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from gdown) (4.12.3)
Requirement already satisfied: filelock in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from gdown) (3.13.4)
Requirement already satisfied: requests[socks] in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from gdown) (2.31.0)
Requirement already satisfied: tqdm in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from gdown) (4.66.2)
Requirement already satisfied: soupsieve>1.2 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from beautifulsoup4->gdown) (2.5)
Requirement already satisfied: charset-normalizer<4,>=2 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from requests[socks]->gdown) (3.3.2)
Requirement already satisfied: idna<4,>=2.5 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from requests[socks]->gdown) (3.7)
Requirement already satisfied: urllib3<3,>=1.21.1 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from requests[socks]->gdown) (2.2.1)
Requirement already satisfied: certifi>=2017.4.17 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from requests[socks]->gdown) (2024.2.2)
Requirement already satisfied: PySocks!=1.5.7,>=1.5.6 in /home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages (from requests[socks]->gdown) (1.7.1)
Downloading...
From (original): https://drive.google.com/uc?id=1llJ4NGNTjY187RLX4MtlmHYhfGxBNWmd
From (redirected): https://drive.google.com/uc?id=1llJ4NGNTjY187RLX4MtlmHYhfGxBNWmd&confirm=t&uuid=7d1b1592-17f3-4232-9d3b-02bf8cd726be
To: /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/ai_multi_model_bundle_data.zip
100%|█████████████████████████████████████████| 647M/647M [00:06<00:00, 105MB/s]
Archive: ai_multi_model_bundle_data.zip
inflating: dcm/1-001.dcm
inflating: dcm/1-002.dcm
inflating: dcm/1-003.dcm
inflating: dcm/1-004.dcm
inflating: dcm/1-005.dcm
inflating: dcm/1-006.dcm
inflating: dcm/1-007.dcm
inflating: dcm/1-008.dcm
inflating: dcm/1-009.dcm
inflating: dcm/1-010.dcm
inflating: dcm/1-011.dcm
inflating: dcm/1-012.dcm
inflating: dcm/1-013.dcm
inflating: dcm/1-014.dcm
inflating: dcm/1-015.dcm
inflating: dcm/1-016.dcm
inflating: dcm/1-017.dcm
inflating: dcm/1-018.dcm
inflating: dcm/1-019.dcm
inflating: dcm/1-020.dcm
inflating: dcm/1-021.dcm
inflating: dcm/1-022.dcm
inflating: dcm/1-023.dcm
inflating: dcm/1-024.dcm
inflating: dcm/1-025.dcm
inflating: dcm/1-026.dcm
inflating: dcm/1-027.dcm
inflating: dcm/1-028.dcm
inflating: dcm/1-029.dcm
inflating: dcm/1-030.dcm
inflating: dcm/1-031.dcm
inflating: dcm/1-032.dcm
inflating: dcm/1-033.dcm
inflating: dcm/1-034.dcm
inflating: dcm/1-035.dcm
inflating: dcm/1-036.dcm
inflating: dcm/1-037.dcm
inflating: dcm/1-038.dcm
inflating: dcm/1-039.dcm
inflating: dcm/1-040.dcm
inflating: dcm/1-041.dcm
inflating: dcm/1-042.dcm
inflating: dcm/1-043.dcm
inflating: dcm/1-044.dcm
inflating: dcm/1-045.dcm
inflating: dcm/1-046.dcm
inflating: dcm/1-047.dcm
inflating: dcm/1-048.dcm
inflating: dcm/1-049.dcm
inflating: dcm/1-050.dcm
inflating: dcm/1-051.dcm
inflating: dcm/1-052.dcm
inflating: dcm/1-053.dcm
inflating: dcm/1-054.dcm
inflating: dcm/1-055.dcm
inflating: dcm/1-056.dcm
inflating: dcm/1-057.dcm
inflating: dcm/1-058.dcm
inflating: dcm/1-059.dcm
inflating: dcm/1-060.dcm
inflating: dcm/1-061.dcm
inflating: dcm/1-062.dcm
inflating: dcm/1-063.dcm
inflating: dcm/1-064.dcm
inflating: dcm/1-065.dcm
inflating: dcm/1-066.dcm
inflating: dcm/1-067.dcm
inflating: dcm/1-068.dcm
inflating: dcm/1-069.dcm
inflating: dcm/1-070.dcm
inflating: dcm/1-071.dcm
inflating: dcm/1-072.dcm
inflating: dcm/1-073.dcm
inflating: dcm/1-074.dcm
inflating: dcm/1-075.dcm
inflating: dcm/1-076.dcm
inflating: dcm/1-077.dcm
inflating: dcm/1-078.dcm
inflating: dcm/1-079.dcm
inflating: dcm/1-080.dcm
inflating: dcm/1-081.dcm
inflating: dcm/1-082.dcm
inflating: dcm/1-083.dcm
inflating: dcm/1-084.dcm
inflating: dcm/1-085.dcm
inflating: dcm/1-086.dcm
inflating: dcm/1-087.dcm
inflating: dcm/1-088.dcm
inflating: dcm/1-089.dcm
inflating: dcm/1-090.dcm
inflating: dcm/1-091.dcm
inflating: dcm/1-092.dcm
inflating: dcm/1-093.dcm
inflating: dcm/1-094.dcm
inflating: dcm/1-095.dcm
inflating: dcm/1-096.dcm
inflating: dcm/1-097.dcm
inflating: dcm/1-098.dcm
inflating: dcm/1-099.dcm
inflating: dcm/1-100.dcm
inflating: dcm/1-101.dcm
inflating: dcm/1-102.dcm
inflating: dcm/1-103.dcm
inflating: dcm/1-104.dcm
inflating: dcm/1-105.dcm
inflating: dcm/1-106.dcm
inflating: dcm/1-107.dcm
inflating: dcm/1-108.dcm
inflating: dcm/1-109.dcm
inflating: dcm/1-110.dcm
inflating: dcm/1-111.dcm
inflating: dcm/1-112.dcm
inflating: dcm/1-113.dcm
inflating: dcm/1-114.dcm
inflating: dcm/1-115.dcm
inflating: dcm/1-116.dcm
inflating: dcm/1-117.dcm
inflating: dcm/1-118.dcm
inflating: dcm/1-119.dcm
inflating: dcm/1-120.dcm
inflating: dcm/1-121.dcm
inflating: dcm/1-122.dcm
inflating: dcm/1-123.dcm
inflating: dcm/1-124.dcm
inflating: dcm/1-125.dcm
inflating: dcm/1-126.dcm
inflating: dcm/1-127.dcm
inflating: dcm/1-128.dcm
inflating: dcm/1-129.dcm
inflating: dcm/1-130.dcm
inflating: dcm/1-131.dcm
inflating: dcm/1-132.dcm
inflating: dcm/1-133.dcm
inflating: dcm/1-134.dcm
inflating: dcm/1-135.dcm
inflating: dcm/1-136.dcm
inflating: dcm/1-137.dcm
inflating: dcm/1-138.dcm
inflating: dcm/1-139.dcm
inflating: dcm/1-140.dcm
inflating: dcm/1-141.dcm
inflating: dcm/1-142.dcm
inflating: dcm/1-143.dcm
inflating: dcm/1-144.dcm
inflating: dcm/1-145.dcm
inflating: dcm/1-146.dcm
inflating: dcm/1-147.dcm
inflating: dcm/1-148.dcm
inflating: dcm/1-149.dcm
inflating: dcm/1-150.dcm
inflating: dcm/1-151.dcm
inflating: dcm/1-152.dcm
inflating: dcm/1-153.dcm
inflating: dcm/1-154.dcm
inflating: dcm/1-155.dcm
inflating: dcm/1-156.dcm
inflating: dcm/1-157.dcm
inflating: dcm/1-158.dcm
inflating: dcm/1-159.dcm
inflating: dcm/1-160.dcm
inflating: dcm/1-161.dcm
inflating: dcm/1-162.dcm
inflating: dcm/1-163.dcm
inflating: dcm/1-164.dcm
inflating: dcm/1-165.dcm
inflating: dcm/1-166.dcm
inflating: dcm/1-167.dcm
inflating: dcm/1-168.dcm
inflating: dcm/1-169.dcm
inflating: dcm/1-170.dcm
inflating: dcm/1-171.dcm
inflating: dcm/1-172.dcm
inflating: dcm/1-173.dcm
inflating: dcm/1-174.dcm
inflating: dcm/1-175.dcm
inflating: dcm/1-176.dcm
inflating: dcm/1-177.dcm
inflating: dcm/1-178.dcm
inflating: dcm/1-179.dcm
inflating: dcm/1-180.dcm
inflating: dcm/1-181.dcm
inflating: dcm/1-182.dcm
inflating: dcm/1-183.dcm
inflating: dcm/1-184.dcm
inflating: dcm/1-185.dcm
inflating: dcm/1-186.dcm
inflating: dcm/1-187.dcm
inflating: dcm/1-188.dcm
inflating: dcm/1-189.dcm
inflating: dcm/1-190.dcm
inflating: dcm/1-191.dcm
inflating: dcm/1-192.dcm
inflating: dcm/1-193.dcm
inflating: dcm/1-194.dcm
inflating: dcm/1-195.dcm
inflating: dcm/1-196.dcm
inflating: dcm/1-197.dcm
inflating: dcm/1-198.dcm
inflating: dcm/1-199.dcm
inflating: dcm/1-200.dcm
inflating: dcm/1-201.dcm
inflating: dcm/1-202.dcm
inflating: dcm/1-203.dcm
inflating: dcm/1-204.dcm
inflating: multi_models/pancreas_ct_dints/model.ts
inflating: multi_models/spleen_ct/model.ts
Set up environment variables¶
models_folder = "multi_models"
%env HOLOSCAN_INPUT_PATH dcm
%env HOLOSCAN_MODEL_PATH {models_folder}
%env HOLOSCAN_OUTPUT_PATH output
env: HOLOSCAN_INPUT_PATH=dcm
env: HOLOSCAN_MODEL_PATH=multi_models
env: HOLOSCAN_OUTPUT_PATH=output
Set up imports¶
Let’s import necessary classes/decorators to define Application and Operator.
import logging
from pathlib import Path
# Required for setting SegmentDescription attributes. Direct import as this is not part of App SDK package.
from pydicom.sr.codedict import codes
from monai.deploy.conditions import CountCondition
from monai.deploy.core import AppContext, Application
from monai.deploy.core.domain import Image
from monai.deploy.core.io_type import IOType
from monai.deploy.operators.dicom_data_loader_operator import DICOMDataLoaderOperator
from monai.deploy.operators.dicom_seg_writer_operator import DICOMSegmentationWriterOperator, SegmentDescription
from monai.deploy.operators.dicom_series_selector_operator import DICOMSeriesSelectorOperator
from monai.deploy.operators.dicom_series_to_volume_operator import DICOMSeriesToVolumeOperator
from monai.deploy.operators.monai_bundle_inference_operator import (
BundleConfigNames,
IOMapping,
MonaiBundleInferenceOperator,
)
from monai.deploy.operators.stl_conversion_operator import STLConversionOperator
Determining the Input and Output for the Model Bundle Inference Operator¶
The App SDK provides a MonaiBundleInferenceOperator
class to perform inference with a MONAI Bundle, which is essentially a PyTorch model in TorchScript with additional metadata describing the model network and processing specification. This operator uses the MONAI utilities to parse a MONAI Bundle to automatically instantiate the objects required for input and output processing as well as inference, as such it depends on MONAI transforms, inferers, and in turn their dependencies.
Each Operator class inherits from the base Operator
base class, and its input/output properties are specified in the setup
function (as opposed to using decorators @input
and @output
in Version 0.5 and below).
For the MonaiBundleInferenceOperator
class, the input/output need to be defined to match those of the model network, both in name and data type. For the current release, an IOMapping
object is used to connect the operator input/output to those of the model network by using the same names. This is likely to change, to be automated, in the future releases once certain limitation in the App SDK is removed.
The Spleen CT Segmentation model network has a named input, called “image”, and the named output called “pred”, and both are of image type, which can all be mapped to the App SDK Image. This piece of information is typically acquired by examining the model metadata network_data_format
attribute in the bundle, as seen in this [example] (https://github.com/Project-MONAI/model-zoo/blob/dev/models/spleen_ct_segmentation/configs/metadata.json).
Creating Application class¶
Our application class would look like below.
It defines App
class, inheriting the base Application
class.
Objects required for DICOM parsing, series selection, pixel data conversion to volume image, model specific inference, and the AI result specific DICOM Segmentation object writers are created. The execution pipeline, as a Directed Acyclic Graph, is then constructed by connecting these objects through self.add_flow()
.
class App(Application):
"""This example demonstrates how to create a multi-model/multi-AI application.
The important steps are:
1. Place the model TorchScripts in a defined folder structure, see below for details
2. Pass the model name to the inference operator instance in the app
3. Connect the input to and output from the inference operators, as required by the app
Required Model Folder Structure:
1. The model TorchScripts, be it MONAI Bundle compliant or not, must be placed in
a parent folder, whose path is used as the path to the model(s) on app execution
2. Each TorchScript file needs to be in a sub-folder, whose name is the model name
An example is shown below, where the `parent_foler` name can be the app's own choosing, and
the sub-folder names become model names, `pancreas_ct_dints` and `spleen_model`, respectively.
<parent_fodler>
├── pancreas_ct_dints
│ └── model.ts
└── spleen_ct
└── model.ts
Note:
1. The TorchScript files of MONAI Bundles can be downloaded from MONAI Model Zoo, at
https://github.com/Project-MONAI/model-zoo/tree/dev/models
https://github.com/Project-MONAI/model-zoo/tree/dev/models/spleen_ct_segmentation, v0.3.2
https://github.com/Project-MONAI/model-zoo/tree/dev/models/pancreas_ct_dints_segmentation, v0.3.8
2. The input DICOM instances are from a DICOM Series of CT Abdomen, similar to the ones
used in the Spleen Segmentation example
3. This example is purely for technical demonstration, not for clinical use
Execution Time Estimate:
With a Nvidia GV100 32GB GPU, the execution time is around 87 seconds for an input DICOM series of 204 instances,
and 167 second for a series of 515 instances.
"""
def __init__(self, *args, **kwargs):
"""Creates an application instance."""
self._logger = logging.getLogger("{}.{}".format(__name__, type(self).__name__))
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
# This method calls the base class to run. Can be omitted if simply calling through.
self._logger.info(f"Begin {self.run.__name__}")
super().run(*args, **kwargs)
self._logger.info(f"End {self.run.__name__}")
def compose(self):
"""Creates the app specific operators and chain them up in the processing DAG."""
logging.info(f"Begin {self.compose.__name__}")
app_context = Application.init_app_context({}) # Do not pass argv in Jupyter Notebook
app_input_path = Path(app_context.input_path)
app_output_path = Path(app_context.output_path)
# Create the custom operator(s) as well as SDK built-in operator(s).
study_loader_op = DICOMDataLoaderOperator(
self, CountCondition(self, 1), input_folder=app_input_path, name="study_loader_op"
)
series_selector_op = DICOMSeriesSelectorOperator(self, rules=Sample_Rules_Text, name="series_selector_op")
series_to_vol_op = DICOMSeriesToVolumeOperator(self, name="series_to_vol_op")
# Create the inference operator that supports MONAI Bundle and automates the inference.
# The IOMapping labels match the input and prediction keys in the pre and post processing.
# The model_name needs to be provided as this is a multi-model application and each inference
# operator need to rely on the name to access the named loaded model network.
# create an inference operator for each.
#
# Pertinent MONAI Bundle:
# https://github.com/Project-MONAI/model-zoo/tree/dev/models/spleen_ct_segmentation, v0.3.2
# https://github.com/Project-MONAI/model-zoo/tree/dev/models/pancreas_ct_dints_segmentation, v0.3.8
config_names = BundleConfigNames(config_names=["inference"]) # Same as the default
# This is the inference operator for the spleen_model bundle. Note the model name.
bundle_spleen_seg_op = MonaiBundleInferenceOperator(
self,
input_mapping=[IOMapping("image", Image, IOType.IN_MEMORY)],
output_mapping=[IOMapping("pred", Image, IOType.IN_MEMORY)],
app_context=app_context,
bundle_config_names=config_names,
model_name="spleen_ct",
name="bundle_spleen_seg_op",
)
# This is the inference operator for the pancreas_ct_dints bundle. Note the model name.
bundle_pancreas_seg_op = MonaiBundleInferenceOperator(
self,
input_mapping=[IOMapping("image", Image, IOType.IN_MEMORY)],
output_mapping=[IOMapping("pred", Image, IOType.IN_MEMORY)],
app_context=app_context,
bundle_config_names=config_names,
model_name="pancreas_ct_dints",
name="bundle_pancreas_seg_op",
)
# Create DICOM Seg writer providing the required segment description for each segment with
# the actual algorithm and the pertinent organ/tissue. The segment_label, algorithm_name,
# and algorithm_version are of DICOM VR LO type, limited to 64 chars.
# https://dicom.nema.org/medical/dicom/current/output/chtml/part05/sect_6.2.html
#
# NOTE: Each generated DICOM Seg will be a dcm file with the name based on the SOP instance UID.
# Description for the Spleen seg, and the seg writer obj
seg_descriptions_spleen = [
SegmentDescription(
segment_label="Spleen",
segmented_property_category=codes.SCT.Organ,
segmented_property_type=codes.SCT.Spleen,
algorithm_name="volumetric (3D) segmentation of the spleen from CT image",
algorithm_family=codes.DCM.ArtificialIntelligence,
algorithm_version="0.3.2",
)
]
custom_tags_spleen = {"SeriesDescription": "AI Spleen Seg for research use only. Not for clinical use."}
dicom_seg_writer_spleen = DICOMSegmentationWriterOperator(
self,
segment_descriptions=seg_descriptions_spleen,
custom_tags=custom_tags_spleen,
output_folder=app_output_path,
name="dicom_seg_writer_spleen",
)
# Description for the Pancreas seg, and the seg writer obj
_algorithm_name = "Pancreas CT DiNTS segmentation from CT image"
_algorithm_family = codes.DCM.ArtificialIntelligence
_algorithm_version = "0.3.8"
seg_descriptions_pancreas = [
SegmentDescription(
segment_label="Pancreas",
segmented_property_category=codes.SCT.Organ,
segmented_property_type=codes.SCT.Pancreas,
algorithm_name=_algorithm_name,
algorithm_family=_algorithm_family,
algorithm_version=_algorithm_version,
),
SegmentDescription(
segment_label="Tumor",
segmented_property_category=codes.SCT.Tumor,
segmented_property_type=codes.SCT.Tumor,
algorithm_name=_algorithm_name,
algorithm_family=_algorithm_family,
algorithm_version=_algorithm_version,
),
]
custom_tags_pancreas = {"SeriesDescription": "AI Pancreas Seg for research use only. Not for clinical use."}
dicom_seg_writer_pancreas = DICOMSegmentationWriterOperator(
self,
segment_descriptions=seg_descriptions_pancreas,
custom_tags=custom_tags_pancreas,
output_folder=app_output_path,
name="dicom_seg_writer_pancreas",
)
# NOTE: Sharp eyed readers can already see that the above instantiation of object can be simply parameterized.
# Very true, but leaving them as if for easy reading. In fact the whole app can be parameterized for general use.
# Create the processing pipeline, by specifying the upstream and downstream operators, and
# ensuring the output from the former matches the input of the latter, in both name and type.
self.add_flow(study_loader_op, series_selector_op, {("dicom_study_list", "dicom_study_list")})
self.add_flow(
series_selector_op, series_to_vol_op, {("study_selected_series_list", "study_selected_series_list")}
)
# Feed the input image to all inference operators
self.add_flow(series_to_vol_op, bundle_spleen_seg_op, {("image", "image")})
# The Pancreas CT Seg bundle requires PyTorch 1.12.0 to avoid failure to load.
self.add_flow(series_to_vol_op, bundle_pancreas_seg_op, {("image", "image")})
# Create DICOM Seg for one of the inference output
# Note below the dicom_seg_writer requires two inputs, each coming from a upstream operator.
self.add_flow(
series_selector_op, dicom_seg_writer_spleen, {("study_selected_series_list", "study_selected_series_list")}
)
self.add_flow(bundle_spleen_seg_op, dicom_seg_writer_spleen, {("pred", "seg_image")})
# Create DICOM Seg for one of the inference output
# Note below the dicom_seg_writer requires two inputs, each coming from a upstream operator.
self.add_flow(
series_selector_op,
dicom_seg_writer_pancreas,
{("study_selected_series_list", "study_selected_series_list")},
)
self.add_flow(bundle_pancreas_seg_op, dicom_seg_writer_pancreas, {("pred", "seg_image")})
logging.info(f"End {self.compose.__name__}")
# This is a sample series selection rule in JSON, simply selecting CT series.
# If the study has more than 1 CT series, then all of them will be selected.
# Please see more detail in DICOMSeriesSelectorOperator.
Sample_Rules_Text = """
{
"selections": [
{
"name": "CT Series",
"conditions": {
"StudyDescription": "(.*?)",
"Modality": "(?i)CT",
"SeriesDescription": "(.*?)"
}
}
]
}
"""
Executing app locally¶
We can execute the app in the Jupyter notebook. Note that the DICOM files of the CT Abdomen series must be present in the input folder, the models are already staged, and environment variables are set.
!rm -rf $HOLOSCAN_OUTPUT_PATH
app = App().run()
[2024-04-23 16:11:04,518] [INFO] (root) - Parsed args: Namespace(log_level=None, input=None, output=None, model=None, workdir=None, argv=[])
[2024-04-23 16:11:04,535] [INFO] (root) - AppContext object: AppContext(input_path=dcm, output_path=output, model_path=multi_models, workdir=)
[2024-04-23 16:11:04,542] [INFO] (root) - End compose
[info] [gxf_executor.cpp:247] Creating context
[info] [gxf_executor.cpp:1672] Loading extensions from configs...
[info] [gxf_executor.cpp:1842] Activating Graph...
[info] [gxf_executor.cpp:1874] Running Graph...
[info] [gxf_executor.cpp:1876] Waiting for completion...
[2024-04-23 16:11:04,579] [INFO] (monai.deploy.operators.dicom_data_loader_operator.DICOMDataLoaderOperator) - No or invalid input path from the optional input port: None
2024-04-23 16:11:04.577 INFO gxf/std/greedy_scheduler.cpp@191: Scheduling 9 entities
[2024-04-23 16:11:05,124] [INFO] (root) - Finding series for Selection named: CT Series
[2024-04-23 16:11:05,125] [INFO] (root) - Searching study, : 1.3.6.1.4.1.14519.5.2.1.7085.2626.822645453932810382886582736291
# of series: 1
[2024-04-23 16:11:05,126] [INFO] (root) - Working on series, instance UID: 1.3.6.1.4.1.14519.5.2.1.7085.2626.119403521930927333027265674239
[2024-04-23 16:11:05,128] [INFO] (root) - On attribute: 'StudyDescription' to match value: '(.*?)'
[2024-04-23 16:11:05,129] [INFO] (root) - Series attribute StudyDescription value: CT ABDOMEN W IV CONTRAST
[2024-04-23 16:11:05,130] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 16:11:05,131] [INFO] (root) - On attribute: 'Modality' to match value: '(?i)CT'
[2024-04-23 16:11:05,132] [INFO] (root) - Series attribute Modality value: CT
[2024-04-23 16:11:05,134] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 16:11:05,136] [INFO] (root) - On attribute: 'SeriesDescription' to match value: '(.*?)'
[2024-04-23 16:11:05,136] [INFO] (root) - Series attribute SeriesDescription value: ABD/PANC 3.0 B31f
[2024-04-23 16:11:05,137] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 16:11:05,137] [INFO] (root) - Selected Series, UID: 1.3.6.1.4.1.14519.5.2.1.7085.2626.119403521930927333027265674239
[2024-04-23 16:11:05,373] [INFO] (root) - Parsing from bundle_path: /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models/pancreas_ct_dints/model.ts
[2024-04-23 16:12:44,188] [INFO] (root) - Parsing from bundle_path: /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models/spleen_ct/model.ts
/home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages/highdicom/valuerep.py:54: UserWarning: The string "C3N-00198" is unlikely to represent the intended person name since it contains only a single component. Construct a person name according to the format in described in https://dicom.nema.org/dicom/2013/output/chtml/part05/sect_6.2.html#sect_6.2.1.2, or, in pydicom 2.2.0 or later, use the pydicom.valuerep.PersonName.from_named_components() method to construct the person name correctly. If a single-component name is really intended, add a trailing caret character to disambiguate the name.
warnings.warn(
[2024-04-23 16:12:49,664] [INFO] (highdicom.base) - copy Image-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:12:49,665] [INFO] (highdicom.base) - copy attributes of module "Specimen"
[2024-04-23 16:12:49,666] [INFO] (highdicom.base) - copy Patient-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:12:49,667] [INFO] (highdicom.base) - copy attributes of module "Patient"
[2024-04-23 16:12:49,669] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Subject"
[2024-04-23 16:12:49,670] [INFO] (highdicom.base) - copy Study-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:12:49,672] [INFO] (highdicom.base) - copy attributes of module "General Study"
[2024-04-23 16:12:49,673] [INFO] (highdicom.base) - copy attributes of module "Patient Study"
[2024-04-23 16:12:49,675] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Study"
[2024-04-23 16:12:50,843] [INFO] (highdicom.base) - copy Image-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:12:50,844] [INFO] (highdicom.base) - copy attributes of module "Specimen"
[2024-04-23 16:12:50,845] [INFO] (highdicom.base) - copy Patient-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:12:50,846] [INFO] (highdicom.base) - copy attributes of module "Patient"
[2024-04-23 16:12:50,847] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Subject"
[2024-04-23 16:12:50,848] [INFO] (highdicom.base) - copy Study-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:12:50,849] [INFO] (highdicom.base) - copy attributes of module "General Study"
[2024-04-23 16:12:50,850] [INFO] (highdicom.base) - copy attributes of module "Patient Study"
[2024-04-23 16:12:50,852] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Study"
[info] [gxf_executor.cpp:1879] Deactivating Graph...
[info] [gxf_executor.cpp:1887] Graph execution finished.
[2024-04-23 16:12:50,954] [INFO] (__main__.App) - End run
2024-04-23 16:12:50.952 INFO gxf/std/greedy_scheduler.cpp@372: Scheduler stopped: Some entities are waiting for execution, but there are no periodic or async entities to get out of the deadlock.
2024-04-23 16:12:50.952 INFO gxf/std/greedy_scheduler.cpp@401: Scheduler finished.
Once the application is verified inside Jupyter notebook, we can write the whole application as a file, adding the following lines:
if __name__ == "__main__":
App().run()
The above lines are needed to execute the application code by using python
interpreter.
A __main__.py
file should also be added, so the application folder structure would look like below:
my_app
├── __main__.py
└── app.py
# Create an application folder
!mkdir -p my_app && rm -rf my_app/*
app.py¶
%%writefile my_app/app.py
# Copyright 2021-2023 MONAI Consortium
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
from pathlib import Path
# Required for setting SegmentDescription attributes. Direct import as this is not part of App SDK package.
from pydicom.sr.codedict import codes
from monai.deploy.conditions import CountCondition
from monai.deploy.core import AppContext, Application
from monai.deploy.core.domain import Image
from monai.deploy.core.io_type import IOType
from monai.deploy.operators.dicom_data_loader_operator import DICOMDataLoaderOperator
from monai.deploy.operators.dicom_seg_writer_operator import DICOMSegmentationWriterOperator, SegmentDescription
from monai.deploy.operators.dicom_series_selector_operator import DICOMSeriesSelectorOperator
from monai.deploy.operators.dicom_series_to_volume_operator import DICOMSeriesToVolumeOperator
from monai.deploy.operators.monai_bundle_inference_operator import (
BundleConfigNames,
IOMapping,
MonaiBundleInferenceOperator,
)
class App(Application):
"""This example demonstrates how to create a multi-model/multi-AI application.
The important steps are:
1. Place the model TorchScripts in a defined folder structure, see below for details
2. Pass the model name to the inference operator instance in the app
3. Connect the input to and output from the inference operators, as required by the app
Required Model Folder Structure:
1. The model TorchScripts, be it MONAI Bundle compliant or not, must be placed in
a parent folder, whose path is used as the path to the model(s) on app execution
2. Each TorchScript file needs to be in a sub-folder, whose name is the model name
An example is shown below, where the `parent_foler` name can be the app's own choosing, and
the sub-folder names become model names, `pancreas_ct_dints` and `spleen_model`, respectively.
<parent_fodler>
├── pancreas_ct_dints
│ └── model.ts
└── spleen_ct
└── model.ts
Note:
1. The TorchScript files of MONAI Bundles can be downloaded from MONAI Model Zoo, at
https://github.com/Project-MONAI/model-zoo/tree/dev/models
https://github.com/Project-MONAI/model-zoo/tree/dev/models/spleen_ct_segmentation, v0.3.2
https://github.com/Project-MONAI/model-zoo/tree/dev/models/pancreas_ct_dints_segmentation, v0.3.8
2. The input DICOM instances are from a DICOM Series of CT Abdomen, similar to the ones
used in the Spleen Segmentation example
3. This example is purely for technical demonstration, not for clinical use
Execution Time Estimate:
With a Nvidia GV100 32GB GPU, the execution time is around 87 seconds for an input DICOM series of 204 instances,
and 167 second for a series of 515 instances.
"""
def __init__(self, *args, **kwargs):
"""Creates an application instance."""
self._logger = logging.getLogger("{}.{}".format(__name__, type(self).__name__))
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
# This method calls the base class to run. Can be omitted if simply calling through.
self._logger.info(f"Begin {self.run.__name__}")
super().run(*args, **kwargs)
self._logger.info(f"End {self.run.__name__}")
def compose(self):
"""Creates the app specific operators and chain them up in the processing DAG."""
logging.info(f"Begin {self.compose.__name__}")
# Use Commandline options over environment variables to init context.
app_context = Application.init_app_context(self.argv)
app_input_path = Path(app_context.input_path)
app_output_path = Path(app_context.output_path)
# Create the custom operator(s) as well as SDK built-in operator(s).
study_loader_op = DICOMDataLoaderOperator(
self, CountCondition(self, 1), input_folder=app_input_path, name="study_loader_op"
)
series_selector_op = DICOMSeriesSelectorOperator(self, rules=Sample_Rules_Text, name="series_selector_op")
series_to_vol_op = DICOMSeriesToVolumeOperator(self, name="series_to_vol_op")
# Create the inference operator that supports MONAI Bundle and automates the inference.
# The IOMapping labels match the input and prediction keys in the pre and post processing.
# The model_name needs to be provided as this is a multi-model application and each inference
# operator need to rely on the name to access the named loaded model network.
# create an inference operator for each.
#
# Pertinent MONAI Bundle:
# https://github.com/Project-MONAI/model-zoo/tree/dev/models/spleen_ct_segmentation, v0.3.2
# https://github.com/Project-MONAI/model-zoo/tree/dev/models/pancreas_ct_dints_segmentation, v0.3.8
config_names = BundleConfigNames(config_names=["inference"]) # Same as the default
# This is the inference operator for the spleen_model bundle. Note the model name.
bundle_spleen_seg_op = MonaiBundleInferenceOperator(
self,
input_mapping=[IOMapping("image", Image, IOType.IN_MEMORY)],
output_mapping=[IOMapping("pred", Image, IOType.IN_MEMORY)],
app_context=app_context,
bundle_config_names=config_names,
model_name="spleen_ct",
name="bundle_spleen_seg_op",
)
# This is the inference operator for the pancreas_ct_dints bundle. Note the model name.
bundle_pancreas_seg_op = MonaiBundleInferenceOperator(
self,
input_mapping=[IOMapping("image", Image, IOType.IN_MEMORY)],
output_mapping=[IOMapping("pred", Image, IOType.IN_MEMORY)],
app_context=app_context,
bundle_config_names=config_names,
model_name="pancreas_ct_dints",
name="bundle_pancreas_seg_op",
)
# Create DICOM Seg writer providing the required segment description for each segment with
# the actual algorithm and the pertinent organ/tissue. The segment_label, algorithm_name,
# and algorithm_version are of DICOM VR LO type, limited to 64 chars.
# https://dicom.nema.org/medical/dicom/current/output/chtml/part05/sect_6.2.html
#
# NOTE: Each generated DICOM Seg will be a dcm file with the name based on the SOP instance UID.
# Description for the Spleen seg, and the seg writer obj
seg_descriptions_spleen = [
SegmentDescription(
segment_label="Spleen",
segmented_property_category=codes.SCT.Organ,
segmented_property_type=codes.SCT.Spleen,
algorithm_name="volumetric (3D) segmentation of the spleen from CT image",
algorithm_family=codes.DCM.ArtificialIntelligence,
algorithm_version="0.3.2",
)
]
custom_tags_spleen = {"SeriesDescription": "AI Spleen Seg for research use only. Not for clinical use."}
dicom_seg_writer_spleen = DICOMSegmentationWriterOperator(
self,
segment_descriptions=seg_descriptions_spleen,
custom_tags=custom_tags_spleen,
output_folder=app_output_path,
name="dicom_seg_writer_spleen",
)
# Description for the Pancreas seg, and the seg writer obj
_algorithm_name = "Pancreas CT DiNTS segmentation from CT image"
_algorithm_family = codes.DCM.ArtificialIntelligence
_algorithm_version = "0.3.8"
seg_descriptions_pancreas = [
SegmentDescription(
segment_label="Pancreas",
segmented_property_category=codes.SCT.Organ,
segmented_property_type=codes.SCT.Pancreas,
algorithm_name=_algorithm_name,
algorithm_family=_algorithm_family,
algorithm_version=_algorithm_version,
),
SegmentDescription(
segment_label="Tumor",
segmented_property_category=codes.SCT.Tumor,
segmented_property_type=codes.SCT.Tumor,
algorithm_name=_algorithm_name,
algorithm_family=_algorithm_family,
algorithm_version=_algorithm_version,
),
]
custom_tags_pancreas = {"SeriesDescription": "AI Pancreas Seg for research use only. Not for clinical use."}
dicom_seg_writer_pancreas = DICOMSegmentationWriterOperator(
self,
segment_descriptions=seg_descriptions_pancreas,
custom_tags=custom_tags_pancreas,
output_folder=app_output_path,
name="dicom_seg_writer_pancreas",
)
# NOTE: Sharp eyed readers can already see that the above instantiation of object can be simply parameterized.
# Very true, but leaving them as if for easy reading. In fact the whole app can be parameterized for general use.
# Create the processing pipeline, by specifying the upstream and downstream operators, and
# ensuring the output from the former matches the input of the latter, in both name and type.
self.add_flow(study_loader_op, series_selector_op, {("dicom_study_list", "dicom_study_list")})
self.add_flow(
series_selector_op, series_to_vol_op, {("study_selected_series_list", "study_selected_series_list")}
)
# Feed the input image to all inference operators
self.add_flow(series_to_vol_op, bundle_spleen_seg_op, {("image", "image")})
# The Pancreas CT Seg bundle requires PyTorch 1.12.0 to avoid failure to load.
self.add_flow(series_to_vol_op, bundle_pancreas_seg_op, {("image", "image")})
# Create DICOM Seg for one of the inference output
# Note below the dicom_seg_writer requires two inputs, each coming from a upstream operator.
self.add_flow(
series_selector_op, dicom_seg_writer_spleen, {("study_selected_series_list", "study_selected_series_list")}
)
self.add_flow(bundle_spleen_seg_op, dicom_seg_writer_spleen, {("pred", "seg_image")})
# Create DICOM Seg for one of the inference output
# Note below the dicom_seg_writer requires two inputs, each coming from a upstream operator.
self.add_flow(
series_selector_op,
dicom_seg_writer_pancreas,
{("study_selected_series_list", "study_selected_series_list")},
)
self.add_flow(bundle_pancreas_seg_op, dicom_seg_writer_pancreas, {("pred", "seg_image")})
logging.info(f"End {self.compose.__name__}")
# This is a sample series selection rule in JSON, simply selecting CT series.
# If the study has more than 1 CT series, then all of them will be selected.
# Please see more detail in DICOMSeriesSelectorOperator.
Sample_Rules_Text = """
{
"selections": [
{
"name": "CT Series",
"conditions": {
"StudyDescription": "(.*?)",
"Modality": "(?i)CT",
"SeriesDescription": "(.*?)"
}
}
]
}
"""
if __name__ == "__main__":
logging.info(f"Begin {__name__}")
App().run()
logging.info(f"End {__name__}")
Writing my_app/app.py
%%writefile my_app/__main__.py
from app import App
if __name__ == "__main__":
App().run()
Writing my_app/__main__.py
!ls my_app
app.py __main__.py
At this time, let’s execute the app on the command line. Note the required e.
Note
Since the environment variables have been set with the specific input data and model paths from earlier steps, it is not necessary to provide the command line options on running the application.
!rm -rf $HOLOSCAN_OUTPUT_PATH
!python my_app
[2024-04-23 16:12:55,648] [INFO] (root) - Parsed args: Namespace(log_level=None, input=None, output=None, model=None, workdir=None, argv=['my_app'])
[2024-04-23 16:12:55,653] [INFO] (root) - AppContext object: AppContext(input_path=dcm, output_path=output, model_path=multi_models, workdir=)
[2024-04-23 16:12:55,655] [INFO] (root) - End compose
[info] [gxf_executor.cpp:247] Creating context
[info] [gxf_executor.cpp:1672] Loading extensions from configs...
[info] [gxf_executor.cpp:1842] Activating Graph...
[info] [gxf_executor.cpp:1874] Running Graph...
[info] [gxf_executor.cpp:1876] Waiting for completion...
2024-04-23 16:12:55.684 INFO gxf/std/greedy_scheduler.cpp@191: Scheduling 9 entities
[2024-04-23 16:12:55,685] [INFO] (monai.deploy.operators.dicom_data_loader_operator.DICOMDataLoaderOperator) - No or invalid input path from the optional input port: None
[2024-04-23 16:12:56,030] [INFO] (root) - Finding series for Selection named: CT Series
[2024-04-23 16:12:56,030] [INFO] (root) - Searching study, : 1.3.6.1.4.1.14519.5.2.1.7085.2626.822645453932810382886582736291
# of series: 1
[2024-04-23 16:12:56,030] [INFO] (root) - Working on series, instance UID: 1.3.6.1.4.1.14519.5.2.1.7085.2626.119403521930927333027265674239
[2024-04-23 16:12:56,030] [INFO] (root) - On attribute: 'StudyDescription' to match value: '(.*?)'
[2024-04-23 16:12:56,030] [INFO] (root) - Series attribute StudyDescription value: CT ABDOMEN W IV CONTRAST
[2024-04-23 16:12:56,030] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 16:12:56,031] [INFO] (root) - On attribute: 'Modality' to match value: '(?i)CT'
[2024-04-23 16:12:56,031] [INFO] (root) - Series attribute Modality value: CT
[2024-04-23 16:12:56,031] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 16:12:56,031] [INFO] (root) - On attribute: 'SeriesDescription' to match value: '(.*?)'
[2024-04-23 16:12:56,031] [INFO] (root) - Series attribute SeriesDescription value: ABD/PANC 3.0 B31f
[2024-04-23 16:12:56,031] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 16:12:56,031] [INFO] (root) - Selected Series, UID: 1.3.6.1.4.1.14519.5.2.1.7085.2626.119403521930927333027265674239
[2024-04-23 16:12:56,243] [INFO] (root) - Parsing from bundle_path: /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models/pancreas_ct_dints/model.ts
[2024-04-23 16:14:36,645] [INFO] (root) - Parsing from bundle_path: /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models/spleen_ct/model.ts
/home/mqin/src/monai-deploy-app-sdk/.venv/lib/python3.10/site-packages/highdicom/valuerep.py:54: UserWarning: The string "C3N-00198" is unlikely to represent the intended person name since it contains only a single component. Construct a person name according to the format in described in https://dicom.nema.org/dicom/2013/output/chtml/part05/sect_6.2.html#sect_6.2.1.2, or, in pydicom 2.2.0 or later, use the pydicom.valuerep.PersonName.from_named_components() method to construct the person name correctly. If a single-component name is really intended, add a trailing caret character to disambiguate the name.
warnings.warn(
[2024-04-23 16:14:42,011] [INFO] (highdicom.base) - copy Image-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:14:42,011] [INFO] (highdicom.base) - copy attributes of module "Specimen"
[2024-04-23 16:14:42,011] [INFO] (highdicom.base) - copy Patient-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:14:42,011] [INFO] (highdicom.base) - copy attributes of module "Patient"
[2024-04-23 16:14:42,012] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Subject"
[2024-04-23 16:14:42,012] [INFO] (highdicom.base) - copy Study-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:14:42,012] [INFO] (highdicom.base) - copy attributes of module "General Study"
[2024-04-23 16:14:42,012] [INFO] (highdicom.base) - copy attributes of module "Patient Study"
[2024-04-23 16:14:42,012] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Study"
[2024-04-23 16:14:43,063] [INFO] (highdicom.base) - copy Image-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:14:43,063] [INFO] (highdicom.base) - copy attributes of module "Specimen"
[2024-04-23 16:14:43,064] [INFO] (highdicom.base) - copy Patient-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:14:43,064] [INFO] (highdicom.base) - copy attributes of module "Patient"
[2024-04-23 16:14:43,064] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Subject"
[2024-04-23 16:14:43,064] [INFO] (highdicom.base) - copy Study-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 16:14:43,064] [INFO] (highdicom.base) - copy attributes of module "General Study"
[2024-04-23 16:14:43,064] [INFO] (highdicom.base) - copy attributes of module "Patient Study"
[2024-04-23 16:14:43,065] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Study"
2024-04-23 16:14:43.154 INFO gxf/std/greedy_scheduler.cpp@372: Scheduler stopped: Some entities are waiting for execution, but there are no periodic or async entities to get out of the deadlock.
2024-04-23 16:14:43.155 INFO gxf/std/greedy_scheduler.cpp@401: Scheduler finished.
[info] [gxf_executor.cpp:1879] Deactivating Graph...
[info] [gxf_executor.cpp:1887] Graph execution finished.
[2024-04-23 16:14:43,155] [INFO] (app.App) - End run
!ls $HOLOSCAN_OUTPUT_PATH
1.2.826.0.1.3680043.10.511.3.10288134553125230635901121822410318.dcm
1.2.826.0.1.3680043.10.511.3.96971936458025058184826371322949123.dcm
Packaging app¶
Let’s package the app with MONAI Application Packager.
In this version of the App SDK, we need to write out the configuration yaml file as well as the package requirements file, in the application folder.
%%writefile my_app/app.yaml
%YAML 1.2
---
application:
title: MONAI Deploy App Package - Multi Model App
version: 1.0
inputFormats: ["file"]
outputFormats: ["file"]
resources:
cpu: 1
gpu: 1
memory: 1Gi
gpuMemory: 10Gi
Writing my_app/app.yaml
%%writefile my_app/requirements.txt
highdicom>=0.18.2
monai>=1.0
nibabel>=3.2.1
numpy>=1.21.6
pydicom>=2.3.0
setuptools>=59.5.0 # for pkg_resources
SimpleITK>=2.0.0
torch>=1.12.0
Writing my_app/requirements.txt
Now we can use the CLI package command to build the MONAI Application Package (MAP) container image based on a supported base image.
Note
Building a MONAI Application Package (Docker image) can take time. Use -l DEBUG
option to see the progress.
tag_prefix = "my_app"
!monai-deploy package my_app -m {models_folder} -c my_app/app.yaml -t {tag_prefix}:1.0 --platform x64-workstation -l DEBUG
[2024-04-23 16:14:45,501] [INFO] (common) - Downloading CLI manifest file...
[2024-04-23 16:14:45,792] [DEBUG] (common) - Validating CLI manifest file...
[2024-04-23 16:14:45,794] [INFO] (packager.parameters) - Application: /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/my_app
[2024-04-23 16:14:45,795] [INFO] (packager.parameters) - Detected application type: Python Module
[2024-04-23 16:14:45,795] [INFO] (packager) - Scanning for models in /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models...
[2024-04-23 16:14:45,795] [DEBUG] (packager) - Model spleen_ct=/home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models/spleen_ct added.
[2024-04-23 16:14:45,795] [DEBUG] (packager) - Model pancreas_ct_dints=/home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/multi_models/pancreas_ct_dints added.
[2024-04-23 16:14:45,796] [INFO] (packager) - Reading application configuration from /home/mqin/src/monai-deploy-app-sdk/notebooks/tutorials/my_app/app.yaml...
[2024-04-23 16:14:45,799] [INFO] (packager) - Generating app.json...
[2024-04-23 16:14:45,799] [INFO] (packager) - Generating pkg.json...
[2024-04-23 16:14:45,809] [DEBUG] (common) -
=============== Begin app.json ===============
{
"apiVersion": "1.0.0",
"command": "[\"python3\", \"/opt/holoscan/app\"]",
"environment": {
"HOLOSCAN_APPLICATION": "/opt/holoscan/app",
"HOLOSCAN_INPUT_PATH": "input/",
"HOLOSCAN_OUTPUT_PATH": "output/",
"HOLOSCAN_WORKDIR": "/var/holoscan",
"HOLOSCAN_MODEL_PATH": "/opt/holoscan/models",
"HOLOSCAN_CONFIG_PATH": "/var/holoscan/app.yaml",
"HOLOSCAN_APP_MANIFEST_PATH": "/etc/holoscan/app.json",
"HOLOSCAN_PKG_MANIFEST_PATH": "/etc/holoscan/pkg.json",
"HOLOSCAN_DOCS_PATH": "/opt/holoscan/docs",
"HOLOSCAN_LOGS_PATH": "/var/holoscan/logs"
},
"input": {
"path": "input/",
"formats": null
},
"liveness": null,
"output": {
"path": "output/",
"formats": null
},
"readiness": null,
"sdk": "monai-deploy",
"sdkVersion": "0.5.1",
"timeout": 0,
"version": 1.0,
"workingDirectory": "/var/holoscan"
}
================ End app.json ================
[2024-04-23 16:14:45,810] [DEBUG] (common) -
=============== Begin pkg.json ===============
{
"apiVersion": "1.0.0",
"applicationRoot": "/opt/holoscan/app",
"modelRoot": "/opt/holoscan/models",
"models": {
"spleen_ct": "/opt/holoscan/models/spleen_ct",
"pancreas_ct_dints": "/opt/holoscan/models/pancreas_ct_dints"
},
"resources": {
"cpu": 1,
"gpu": 1,
"memory": "1Gi",
"gpuMemory": "10Gi"
},
"version": 1.0,
"platformConfig": "dgpu"
}
================ End pkg.json ================
[2024-04-23 16:14:46,274] [DEBUG] (packager.builder) -
========== Begin Dockerfile ==========
FROM nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu
ENV DEBIAN_FRONTEND=noninteractive
ENV TERM=xterm-256color
ARG UNAME
ARG UID
ARG GID
RUN mkdir -p /etc/holoscan/ \
&& mkdir -p /opt/holoscan/ \
&& mkdir -p /var/holoscan \
&& mkdir -p /opt/holoscan/app \
&& mkdir -p /var/holoscan/input \
&& mkdir -p /var/holoscan/output
LABEL base="nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu"
LABEL tag="my_app:1.0"
LABEL org.opencontainers.image.title="MONAI Deploy App Package - Multi Model App"
LABEL org.opencontainers.image.version="1.0"
LABEL org.nvidia.holoscan="2.0.0"
LABEL org.monai.deploy.app-sdk="0.5.1"
ENV HOLOSCAN_ENABLE_HEALTH_CHECK=true
ENV HOLOSCAN_INPUT_PATH=/var/holoscan/input
ENV HOLOSCAN_OUTPUT_PATH=/var/holoscan/output
ENV HOLOSCAN_WORKDIR=/var/holoscan
ENV HOLOSCAN_APPLICATION=/opt/holoscan/app
ENV HOLOSCAN_TIMEOUT=0
ENV HOLOSCAN_MODEL_PATH=/opt/holoscan/models
ENV HOLOSCAN_DOCS_PATH=/opt/holoscan/docs
ENV HOLOSCAN_CONFIG_PATH=/var/holoscan/app.yaml
ENV HOLOSCAN_APP_MANIFEST_PATH=/etc/holoscan/app.json
ENV HOLOSCAN_PKG_MANIFEST_PATH=/etc/holoscan/pkg.json
ENV HOLOSCAN_LOGS_PATH=/var/holoscan/logs
ENV PATH=/root/.local/bin:/opt/nvidia/holoscan:$PATH
ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/libtorch/1.13.1/lib/:/opt/nvidia/holoscan/lib
RUN apt-get update \
&& apt-get install -y curl jq \
&& rm -rf /var/lib/apt/lists/*
ENV PYTHONPATH="/opt/holoscan/app:$PYTHONPATH"
RUN groupadd -f -g $GID $UNAME
RUN useradd -rm -d /home/$UNAME -s /bin/bash -g $GID -G sudo -u $UID $UNAME
RUN chown -R holoscan /var/holoscan
RUN chown -R holoscan /var/holoscan/input
RUN chown -R holoscan /var/holoscan/output
# Set the working directory
WORKDIR /var/holoscan
# Copy HAP/MAP tool script
COPY ./tools /var/holoscan/tools
RUN chmod +x /var/holoscan/tools
# Copy gRPC health probe
USER $UNAME
ENV PATH=/root/.local/bin:/home/holoscan/.local/bin:/opt/nvidia/holoscan:$PATH
COPY ./pip/requirements.txt /tmp/requirements.txt
RUN pip install --upgrade pip
RUN pip install --no-cache-dir --user -r /tmp/requirements.txt
# MONAI Deploy
# Copy user-specified MONAI Deploy SDK file
COPY ./monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl /tmp/monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl
RUN pip install /tmp/monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl
COPY ./models /opt/holoscan/models
COPY ./map/app.json /etc/holoscan/app.json
COPY ./app.config /var/holoscan/app.yaml
COPY ./map/pkg.json /etc/holoscan/pkg.json
COPY ./app /opt/holoscan/app
ENTRYPOINT ["/var/holoscan/tools"]
=========== End Dockerfile ===========
[2024-04-23 16:14:46,274] [INFO] (packager.builder) -
===============================================================================
Building image for: x64-workstation
Architecture: linux/amd64
Base Image: nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu
Build Image: N/A
Cache: Enabled
Configuration: dgpu
Holoscan SDK Package: pypi.org
MONAI Deploy App SDK Package: /home/mqin/src/monai-deploy-app-sdk/dist/monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl
gRPC Health Probe: N/A
SDK Version: 2.0.0
SDK: monai-deploy
Tag: my_app-x64-workstation-dgpu-linux-amd64:1.0
[2024-04-23 16:14:46,562] [INFO] (common) - Using existing Docker BuildKit builder `holoscan_app_builder`
[2024-04-23 16:14:46,562] [DEBUG] (packager.builder) - Building Holoscan Application Package: tag=my_app-x64-workstation-dgpu-linux-amd64:1.0
#0 building with "holoscan_app_builder" instance using docker-container driver
#1 [internal] load build definition from Dockerfile
#1 transferring dockerfile: 2.65kB done
#1 DONE 0.0s
#2 [internal] load metadata for nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu
#2 DONE 0.4s
#3 [internal] load .dockerignore
#3 transferring context: 1.79kB done
#3 DONE 0.0s
#4 [internal] load build context
#4 DONE 0.0s
#5 importing cache manifest from local:13557986215550987099
#5 inferred cache manifest type: application/vnd.oci.image.index.v1+json done
#5 DONE 0.0s
#6 [ 1/21] FROM nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu@sha256:20adbccd2c7b12dfb1798f6953f071631c3b85cd337858a7506f8e420add6d4a
#6 resolve nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu@sha256:20adbccd2c7b12dfb1798f6953f071631c3b85cd337858a7506f8e420add6d4a 0.0s done
#6 DONE 0.0s
#7 importing cache manifest from nvcr.io/nvidia/clara-holoscan/holoscan:v2.0.0-dgpu
#7 inferred cache manifest type: application/vnd.docker.distribution.manifest.list.v2+json done
#7 DONE 0.4s
#4 [internal] load build context
#4 transferring context: 636.05MB 3.2s done
#4 DONE 3.2s
#8 [15/21] COPY ./monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl /tmp/monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl
#8 CACHED
#9 [ 6/21] RUN chown -R holoscan /var/holoscan
#9 CACHED
#10 [ 5/21] RUN useradd -rm -d /home/holoscan -s /bin/bash -g 1000 -G sudo -u 1000 holoscan
#10 CACHED
#11 [ 4/21] RUN groupadd -f -g 1000 holoscan
#11 CACHED
#12 [ 3/21] RUN apt-get update && apt-get install -y curl jq && rm -rf /var/lib/apt/lists/*
#12 CACHED
#13 [11/21] RUN chmod +x /var/holoscan/tools
#13 CACHED
#14 [12/21] COPY ./pip/requirements.txt /tmp/requirements.txt
#14 CACHED
#15 [ 7/21] RUN chown -R holoscan /var/holoscan/input
#15 CACHED
#16 [14/21] RUN pip install --no-cache-dir --user -r /tmp/requirements.txt
#16 CACHED
#17 [ 8/21] RUN chown -R holoscan /var/holoscan/output
#17 CACHED
#18 [ 9/21] WORKDIR /var/holoscan
#18 CACHED
#19 [ 2/21] RUN mkdir -p /etc/holoscan/ && mkdir -p /opt/holoscan/ && mkdir -p /var/holoscan && mkdir -p /opt/holoscan/app && mkdir -p /var/holoscan/input && mkdir -p /var/holoscan/output
#19 CACHED
#20 [10/21] COPY ./tools /var/holoscan/tools
#20 CACHED
#21 [13/21] RUN pip install --upgrade pip
#21 CACHED
#22 [16/21] RUN pip install /tmp/monai_deploy_app_sdk-0.5.1+20.gb869749.dirty-py3-none-any.whl
#22 CACHED
#23 [17/21] COPY ./models /opt/holoscan/models
#23 DONE 3.9s
#24 [18/21] COPY ./map/app.json /etc/holoscan/app.json
#24 DONE 0.0s
#25 [19/21] COPY ./app.config /var/holoscan/app.yaml
#25 DONE 0.0s
#26 [20/21] COPY ./map/pkg.json /etc/holoscan/pkg.json
#26 DONE 0.0s
#27 [21/21] COPY ./app /opt/holoscan/app
#27 DONE 0.0s
#28 exporting to docker image format
#28 exporting layers
#28 exporting layers 17.8s done
#28 exporting manifest sha256:26808437a116257ae2799583b42bbf04923e1157f8f98341c3403ee35eb234bf 0.0s done
#28 exporting config sha256:95614919d60e7f30ab5d64a25a5fc25ad64d0b166046c60549b5abae2745be7c 0.0s done
#28 sending tarball
#28 ...
#29 importing to docker
#29 loading layer ecf27683cffd 557.06kB / 584.49MB
#29 loading layer ecf27683cffd 154.86MB / 584.49MB 2.1s
#29 loading layer ecf27683cffd 309.72MB / 584.49MB 4.1s
#29 loading layer ecf27683cffd 464.03MB / 584.49MB 6.2s
#29 loading layer 3202d6efcdaa 493B / 493B
#29 loading layer 84f909517c68 311B / 311B
#29 loading layer 1d1958b729ff 323B / 323B
#29 loading layer 8f69ecd226c9 4.00kB / 4.00kB
#29 loading layer 3202d6efcdaa 493B / 493B 2.6s done
#29 loading layer ecf27683cffd 464.03MB / 584.49MB 10.6s done
#29 loading layer 84f909517c68 311B / 311B 2.2s done
#29 loading layer 1d1958b729ff 323B / 323B 1.8s done
#29 loading layer 8f69ecd226c9 4.00kB / 4.00kB 1.5s done
#29 DONE 10.6s
#28 exporting to docker image format
#28 sending tarball 72.8s done
#28 DONE 90.7s
#30 exporting cache to client directory
#30 preparing build cache for export
#30 writing layer sha256:014cff740c9ec6e9a30d0b859219a700ae880eb385d62095d348f5ea136d6015
#30 writing layer sha256:014cff740c9ec6e9a30d0b859219a700ae880eb385d62095d348f5ea136d6015 done
#30 writing layer sha256:0487800842442c7a031a39e1e1857bc6dae4b4f7e5daf3d625f7a8a4833fb364 done
#30 writing layer sha256:06c6aee94862daf0603783db4e1de6f8524b30ac9fbe0374ab3f1d85b2f76f7f done
#30 writing layer sha256:0a1756432df4a4350712d8ae5c003f1526bd2180800b3ae6301cfc9ccf370254 done
#30 writing layer sha256:0a77dcbd0e648ddc4f8e5230ade8fdb781d99e24fa4f13ca96a360c7f7e6751f done
#30 writing layer sha256:0cbe3b20b9b7d01bcb9770de54ba9a54febb401dc371bbb1d8debf1f9850b356 0.0s done
#30 writing layer sha256:0ec682bf99715a9f88631226f3749e2271b8b9f254528ef61f65ed829984821c done
#30 writing layer sha256:1c5c3aa9c2c8bfd1b9eb36248f5b6d67b3db73ef43440f9dd897615771974b39 done
#30 writing layer sha256:1f4a978bb76db2d138cfe7c7c9e76db4096247b06e34d349a2ed504bcd6a7ead done
#30 writing layer sha256:1f73278b7f17492ce1a8b28b139d54596961596d6790dc20046fa6d5909f3e9c done
#30 writing layer sha256:20d331454f5fb557f2692dfbdbe092c718fd2cb55d5db9d661b62228dacca5c2 done
#30 writing layer sha256:20e14f0a8ca68167afb8296c10d7a1b4c3b17b54681cbf3b9b45e1be96afa699 done
#30 writing layer sha256:238f69a43816e481f0295995fcf5fe74d59facf0f9f99734c8d0a2fb140630e0 done
#30 writing layer sha256:255cc51d2e47738a5db3059cbe9f403785cf9496c7df8a28a3c9f0c46a0b3b58 done
#30 writing layer sha256:29eb21ed7ba7894cd28ee75b082877845b969c6e65d34a16b7dbd8630c38f5fe
#30 writing layer sha256:29eb21ed7ba7894cd28ee75b082877845b969c6e65d34a16b7dbd8630c38f5fe 10.2s done
#30 writing layer sha256:2ad84487f9d4d31cd1e0a92697a5447dd241935253d036b272ef16d31620c1e7
#30 writing layer sha256:2ad84487f9d4d31cd1e0a92697a5447dd241935253d036b272ef16d31620c1e7 done
#30 writing layer sha256:2f65750928993b5b31fe572d9e085b53853c5a344feeb0e8615898e285a8c256 done
#30 writing layer sha256:3777c6498f08c0400339c243e827d465075b7296eb2526e38d9b01c84f8764d8 done
#30 writing layer sha256:3e3e04011ebdba380ab129f0ee390626cb2a600623815ca756340c18bedb9517 done
#30 writing layer sha256:42619ce4a0c9e54cfd0ee41a8e5f27d58b3f51becabd1ac6de725fbe6c42b14a done
#30 writing layer sha256:49bdc9abf8a437ccff67cc11490ba52c976577992909856a86be872a34d3b950 done
#30 writing layer sha256:4b691ba9f48b41eaa0c754feba8366f1c030464fcbc55eeffa6c86675990933a done
#30 writing layer sha256:4d04a8db404f16c2704fa10739cb6745a0187713a21a6ef0deb34b48629b54c1 done
#30 writing layer sha256:4f4fb700ef54461cfa02571ae0db9a0dc1e0cdb5577484a6d75e68dc38e8acc1 done
#30 writing layer sha256:542bc8c8d18fbc95e6794122c3593a4a693f8ab6dda4460406f4d7b1ae64a2bc done
#30 writing layer sha256:57f244836ad318f9bbb3b29856ae1a5b31038bfbb9b43d2466d51c199eb55041 done
#30 writing layer sha256:5b5b131e0f20db4cb8e568b623a95f8fc16ed1c6b322a9366df70b59a881f24f done
#30 writing layer sha256:5b90d17b5048adcadefd0b1e4dba9a99247a8827a887e1ca042df375c85b518d done
#30 writing layer sha256:62452179df7c18e292f141d4aec29e6aba9ff8270c893731169fc6f41dc07631 done
#30 writing layer sha256:6630c387f5f2115bca2e646fd0c2f64e1f3d5431c2e050abe607633883eda230 done
#30 writing layer sha256:6661e0146e77a8bcb03edbfda95bf7780c8bb4c4f98bc03a398c88f4b2403d12 done
#30 writing layer sha256:717ebf8c9c66ae393ad01e50dbac4413d7b026b9c97d4d348b22ad17052a1a35 done
#30 writing layer sha256:773c6815e5e7d6855a62f8c5e2fabce3d939ded36c5420f15b54dd7908cdbcfa done
#30 writing layer sha256:7852b73ea931e3a8d3287ee7ef3cf4bad068e44f046583bfc2b81336fb299284 done
#30 writing layer sha256:7f8ec130348bcdac81c295e37fe82b4a6e5e9a3ca980a6343809c561020d82d7 done
#30 writing layer sha256:80885adcad6b5d021bb9f68b6c952018085bb4ce72011bdc0cf7fe8178b5960b done
#30 writing layer sha256:82a3436133b2b17bb407c7fe488932aa0ca55411f23ab55c34a6134b287c6a27 done
#30 writing layer sha256:8371d15eb4d69b1d98174dd098b8ddd5c4f19ec6f8d8b67e72dfa9891dc454b4 done
#30 writing layer sha256:85713f9b166b5add777c524ee807f6265d88b967cbeb9f961d6b09bf220c9a65 done
#30 writing layer sha256:8fe00505006a09966e763918147ef6ed55bb6695b26e4940c780ee430dc5da8e done
#30 writing layer sha256:90eae6faa5cc5ba62f12c25915cdfb1a7a51abfba0d05cb5818c3f908f4e345f done
#30 writing layer sha256:9205d97d9d3e906698bcc6c42d45727c2fa6ec2622abf953d46778c3b8c78edc done
#30 writing layer sha256:92301d1270c19cab329818fb215b41138720ab9b588a2070107860f0b6fb5e11 done
#30 writing layer sha256:993369dbcc13162a6654d2a3e990b8d8b5f37963564d25710e12764337261ae3 done
#30 writing layer sha256:99e42a4adebadb39bf55bf94bbd9fb8034230ee19b6b0a42e6ff96f2e7794f30 done
#30 writing layer sha256:9ac855545fa90ed2bf3b388fdff9ef06ac9427b0c0fca07c9e59161983d8827e done
#30 writing layer sha256:9d19ee268e0d7bcf6716e6658ee1b0384a71d6f2f9aa1ae2085610cf7c7b316f done
#30 writing layer sha256:9fafbd4203c4fefe007a462e0d2cd4c1c7c41db2cfdc58d212279e1b9b4b230c done
#30 writing layer sha256:a1748eee9d376f97bd19225ba61dfada9986f063f4fc429e435f157abb629fc6 done
#30 writing layer sha256:a251fe5ae6c6d2d5034e4ca88b5dfe5d4827ff90b18e9b143a073232a32bb18d done
#30 writing layer sha256:a68f4e0ec09ec3b78cb4cf8e4511d658e34e7b6f676d7806ad9703194ff17604 done
#30 writing layer sha256:a8e4decc8f7289623b8fd7b9ba1ca555b5a755ebdbf81328d68209f148d9e602 done
#30 writing layer sha256:ac69ec878bf5e3f9294e9726cd393e4133a5ea3934256636de00ddc558f25e55 0.0s done
#30 writing layer sha256:afde1c269453ce68a0f2b54c1ba8c5ecddeb18a19e5618a4acdef1f0fe3921af done
#30 writing layer sha256:b12b2b11632995c7c56700227a9c2f2016dab20c083c140a835c551bb2aed460 0.0s done
#30 writing layer sha256:b406feb20a37b8c87ef4f5ef814039e3adc90473d50c366b7d9bb6ded4e94a2e
#30 preparing build cache for export 11.6s done
#30 writing layer sha256:b406feb20a37b8c87ef4f5ef814039e3adc90473d50c366b7d9bb6ded4e94a2e done
#30 writing layer sha256:b48a5fafcaba74eb5d7e7665601509e2889285b50a04b5b639a23f8adc818157 done
#30 writing layer sha256:ba9f7c75e4dd7942b944679995365aab766d3677da2e69e1d74472f471a484dd done
#30 writing layer sha256:bdc13166216ae226fa6976f9ce91f4f259d43972f1e0a9b723e436919534b2f4 done
#30 writing layer sha256:c815f0be64eded102822d81e029bd23b0d8d9a0fbfeb492ec0b4b0bc4ee777bf done
#30 writing layer sha256:c97f7fb19e2e0b8ee3e1065f4dee369e35029cc620cafb7fe3dec2e9e06a3ae0 done
#30 writing layer sha256:c98533d2908f36a5e9b52faae83809b3b6865b50e90e2817308acfc64cd3655f done
#30 writing layer sha256:d7da5c5e9a40c476c4b3188a845e3276dedfd752e015ea5113df5af64d4d43f7 done
#30 writing layer sha256:db20521a869adda8244cb64b783c65e1a911efaae0e73ae00e4a34ea6213d6ce done
#30 writing layer sha256:df4fd0ac710d7af949afbc6d25b5b4daf3f0596dabf3dec36fa7ca8fa6e1d049 done
#30 writing layer sha256:e291ddecfbe16b95ee9e90b5e90b1a3d0cfd53dc5e720d6b0f3d28e4a47cf5ac done
#30 writing layer sha256:e8acb678f16bc0c369d5cf9c184f2d3a1c773986816526e5e3e9c0354f7e757f done
#30 writing layer sha256:e9225f7ab6606813ec9acba98a064826ebfd6713a9645a58cd068538af1ecddb done
#30 writing layer sha256:e9808a628d8776e25114053fec83adbf9c6dd084ac0c275042fa8b89db830da5 0.0s done
#30 writing layer sha256:f249faf9663a96b0911a903f8803b11a553c59b698013fb8343492fefdaaea90 done
#30 writing layer sha256:f608e2fbff86e98627b7e462057e7d2416522096d73fe4664b82fe6ce8a4047d done
#30 writing layer sha256:f65d191416580d6c38e3d95eee12377b75a4df548be1492618ce2a8c3c41b99e done
#30 writing config sha256:bad5d754263c9813832b5d1594a04b9c78a10052ec0a59ade6ee42de1baadb19 0.0s done
#30 writing cache manifest sha256:62a85925ced740e444a8f19b3c59ade29144854a44168d8786e00e93dbad2f5f 0.0s done
#30 DONE 11.6s
[2024-04-23 16:16:38,598] [INFO] (packager) - Build Summary:
Platform: x64-workstation/dgpu
Status: Succeeded
Docker Tag: my_app-x64-workstation-dgpu-linux-amd64:1.0
Tarball: None
We can see that the Docker image is created.
!docker image ls | grep {tag_prefix}
my_app-x64-workstation-dgpu-linux-amd64 1.0 95614919d60e About a minute ago 18.3GB
We can choose to display and inspect the MAP manifests by running the container with the show
command.
Furthermore, we can also extract the manifests and other contents in the MAP by using the extract
command while mapping specific folder to the host’s (we know that our MAP is compliant and supports these commands).
Note
The host folder for storing the extracted content must first be created by the user, and if it has been created by Docker on running the container, the folder needs to be deleted and re-created.
!echo "Display manifests and extract MAP contents to the host folder, ./export"
!docker run --rm {tag_prefix}-x64-workstation-dgpu-linux-amd64:1.0 show
!rm -rf `pwd`/export && mkdir -p `pwd`/export
!docker run --rm -v `pwd`/export/:/var/run/holoscan/export/ {tag_prefix}-x64-workstation-dgpu-linux-amd64:1.0 extract
!ls `pwd`/export
Display manifests and extract MAP contents to the host folder, ./export
============================== app.json ==============================
{
"apiVersion": "1.0.0",
"command": "[\"python3\", \"/opt/holoscan/app\"]",
"environment": {
"HOLOSCAN_APPLICATION": "/opt/holoscan/app",
"HOLOSCAN_INPUT_PATH": "input/",
"HOLOSCAN_OUTPUT_PATH": "output/",
"HOLOSCAN_WORKDIR": "/var/holoscan",
"HOLOSCAN_MODEL_PATH": "/opt/holoscan/models",
"HOLOSCAN_CONFIG_PATH": "/var/holoscan/app.yaml",
"HOLOSCAN_APP_MANIFEST_PATH": "/etc/holoscan/app.json",
"HOLOSCAN_PKG_MANIFEST_PATH": "/etc/holoscan/pkg.json",
"HOLOSCAN_DOCS_PATH": "/opt/holoscan/docs",
"HOLOSCAN_LOGS_PATH": "/var/holoscan/logs"
},
"input": {
"path": "input/",
"formats": null
},
"liveness": null,
"output": {
"path": "output/",
"formats": null
},
"readiness": null,
"sdk": "monai-deploy",
"sdkVersion": "0.5.1",
"timeout": 0,
"version": 1,
"workingDirectory": "/var/holoscan"
}
============================== pkg.json ==============================
{
"apiVersion": "1.0.0",
"applicationRoot": "/opt/holoscan/app",
"modelRoot": "/opt/holoscan/models",
"models": {
"spleen_ct": "/opt/holoscan/models/spleen_ct",
"pancreas_ct_dints": "/opt/holoscan/models/pancreas_ct_dints"
},
"resources": {
"cpu": 1,
"gpu": 1,
"memory": "1Gi",
"gpuMemory": "10Gi"
},
"version": 1,
"platformConfig": "dgpu"
}
2024-04-23 23:16:41 [INFO] Copying application from /opt/holoscan/app to /var/run/holoscan/export/app
2024-04-23 23:16:41 [INFO] Copying application manifest file from /etc/holoscan/app.json to /var/run/holoscan/export/config/app.json
2024-04-23 23:16:41 [INFO] Copying pkg manifest file from /etc/holoscan/pkg.json to /var/run/holoscan/export/config/pkg.json
2024-04-23 23:16:41 [INFO] Copying application configuration from /var/holoscan/app.yaml to /var/run/holoscan/export/config/app.yaml
2024-04-23 23:16:41 [INFO] Copying models from /opt/holoscan/models to /var/run/holoscan/export/models
2024-04-23 23:16:42 [INFO] Copying documentation from /opt/holoscan/docs/ to /var/run/holoscan/export/docs
2024-04-23 23:16:42 [INFO] '/opt/holoscan/docs/' cannot be found.
app config models
Executing packaged app locally¶
The packaged app can be run locally through MONAI Application Runner.
# Clear the output folder and run the MAP. The input is expected to be a folder.
!rm -rf $HOLOSCAN_OUTPUT_PATH
!monai-deploy run -i $HOLOSCAN_INPUT_PATH -o $HOLOSCAN_OUTPUT_PATH my_app-x64-workstation-dgpu-linux-amd64:1.0
[2024-04-23 16:16:45,664] [INFO] (runner) - Checking dependencies...
[2024-04-23 16:16:45,664] [INFO] (runner) - --> Verifying if "docker" is installed...
[2024-04-23 16:16:45,664] [INFO] (runner) - --> Verifying if "docker-buildx" is installed...
[2024-04-23 16:16:45,664] [INFO] (runner) - --> Verifying if "my_app-x64-workstation-dgpu-linux-amd64:1.0" is available...
[2024-04-23 16:16:45,739] [INFO] (runner) - Reading HAP/MAP manifest...
Preparing to copy...?25lCopying from container - 0B?25hSuccessfully copied 2.56kB to /tmp/tmpymoiz9gd/app.json
Preparing to copy...?25lCopying from container - 0B?25hSuccessfully copied 2.05kB to /tmp/tmpymoiz9gd/pkg.json
[2024-04-23 16:16:46,001] [INFO] (runner) - --> Verifying if "nvidia-ctk" is installed...
[2024-04-23 16:16:46,002] [INFO] (runner) - --> Verifying "nvidia-ctk" version...
[2024-04-23 16:16:46,316] [INFO] (common) - Launching container (42b411558fd7) using image 'my_app-x64-workstation-dgpu-linux-amd64:1.0'...
container name: silly_davinci
host name: mingq-dt
network: host
user: 1000:1000
ulimits: memlock=-1:-1, stack=67108864:67108864
cap_add: CAP_SYS_PTRACE
ipc mode: host
shared memory size: 67108864
devices:
group_add: 44
2024-04-23 23:16:46 [INFO] Launching application python3 /opt/holoscan/app ...
[2024-04-23 23:16:50,090] [INFO] (root) - Parsed args: Namespace(log_level=None, input=None, output=None, model=None, workdir=None, argv=['/opt/holoscan/app'])
[2024-04-23 23:16:50,095] [INFO] (root) - AppContext object: AppContext(input_path=/var/holoscan/input, output_path=/var/holoscan/output, model_path=/opt/holoscan/models, workdir=/var/holoscan)
[2024-04-23 23:16:50,097] [INFO] (root) - End compose
[info] [app_driver.cpp:1161] Launching the driver/health checking service
[info] [gxf_executor.cpp:247] Creating context
[info] [server.cpp:87] Health checking server listening on 0.0.0.0:8777
[info] [gxf_executor.cpp:1672] Loading extensions from configs...
[info] [gxf_executor.cpp:1842] Activating Graph...
[info] [gxf_executor.cpp:1874] Running Graph...
[info] [gxf_executor.cpp:1876] Waiting for completion...
2024-04-23 23:16:50.129 INFO gxf/std/greedy_scheduler.cpp@191: Scheduling 9 entities
[2024-04-23 23:16:50,131] [INFO] (monai.deploy.operators.dicom_data_loader_operator.DICOMDataLoaderOperator) - No or invalid input path from the optional input port: None
[2024-04-23 23:16:50,512] [INFO] (root) - Finding series for Selection named: CT Series
[2024-04-23 23:16:50,512] [INFO] (root) - Searching study, : 1.3.6.1.4.1.14519.5.2.1.7085.2626.822645453932810382886582736291
# of series: 1
[2024-04-23 23:16:50,512] [INFO] (root) - Working on series, instance UID: 1.3.6.1.4.1.14519.5.2.1.7085.2626.119403521930927333027265674239
[2024-04-23 23:16:50,512] [INFO] (root) - On attribute: 'StudyDescription' to match value: '(.*?)'
[2024-04-23 23:16:50,512] [INFO] (root) - Series attribute StudyDescription value: CT ABDOMEN W IV CONTRAST
[2024-04-23 23:16:50,512] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 23:16:50,512] [INFO] (root) - On attribute: 'Modality' to match value: '(?i)CT'
[2024-04-23 23:16:50,512] [INFO] (root) - Series attribute Modality value: CT
[2024-04-23 23:16:50,512] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 23:16:50,513] [INFO] (root) - On attribute: 'SeriesDescription' to match value: '(.*?)'
[2024-04-23 23:16:50,513] [INFO] (root) - Series attribute SeriesDescription value: ABD/PANC 3.0 B31f
[2024-04-23 23:16:50,513] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2024-04-23 23:16:50,513] [INFO] (root) - Selected Series, UID: 1.3.6.1.4.1.14519.5.2.1.7085.2626.119403521930927333027265674239
[2024-04-23 23:16:50,745] [INFO] (root) - Parsing from bundle_path: /opt/holoscan/models/pancreas_ct_dints/model.ts
[2024-04-23 23:18:36,681] [INFO] (root) - Parsing from bundle_path: /opt/holoscan/models/spleen_ct/model.ts
/home/holoscan/.local/lib/python3.10/site-packages/highdicom/valuerep.py:54: UserWarning: The string "C3N-00198" is unlikely to represent the intended person name since it contains only a single component. Construct a person name according to the format in described in https://dicom.nema.org/dicom/2013/output/chtml/part05/sect_6.2.html#sect_6.2.1.2, or, in pydicom 2.2.0 or later, use the pydicom.valuerep.PersonName.from_named_components() method to construct the person name correctly. If a single-component name is really intended, add a trailing caret character to disambiguate the name.
warnings.warn(
[2024-04-23 23:18:40,078] [INFO] (highdicom.base) - copy Image-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 23:18:40,078] [INFO] (highdicom.base) - copy attributes of module "Specimen"
[2024-04-23 23:18:40,078] [INFO] (highdicom.base) - copy Patient-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 23:18:40,078] [INFO] (highdicom.base) - copy attributes of module "Patient"
[2024-04-23 23:18:40,078] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Subject"
[2024-04-23 23:18:40,079] [INFO] (highdicom.base) - copy Study-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 23:18:40,079] [INFO] (highdicom.base) - copy attributes of module "General Study"
[2024-04-23 23:18:40,079] [INFO] (highdicom.base) - copy attributes of module "Patient Study"
[2024-04-23 23:18:40,079] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Study"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy Image-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy attributes of module "Specimen"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy Patient-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy attributes of module "Patient"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Subject"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy Study-related attributes from dataset "1.3.6.1.4.1.14519.5.2.1.7085.2626.936983343951485811186213470191"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy attributes of module "General Study"
[2024-04-23 23:18:41,286] [INFO] (highdicom.base) - copy attributes of module "Patient Study"
[2024-04-23 23:18:41,287] [INFO] (highdicom.base) - copy attributes of module "Clinical Trial Study"
2024-04-23 23:18:41.373 INFO gxf/std/greedy_scheduler.cpp@372: Scheduler stopped: Some entities are waiting for execution, but there are no periodic or async entities to get out of the deadlock.
[info] [gxf_executor.cpp:1879] Deactivating Graph...
2024-04-23 23:18:41.377 INFO gxf/std/greedy_scheduler.cpp@401: Scheduler finished.
[info] [gxf_executor.cpp:1887] Graph execution finished.
[2024-04-23 23:18:41,385] [INFO] (app.App) - End run
[2024-04-23 16:18:42,406] [INFO] (common) - Container 'silly_davinci'(42b411558fd7) exited.
In the output folder are the DICOM segementation files.
!ls $HOLOSCAN_OUTPUT_PATH
1.2.826.0.1.3680043.10.511.3.11468162564679998832192298844993783.dcm
1.2.826.0.1.3680043.10.511.3.22886221740567104559887431846790837.dcm