9.14. Chest X-ray Classification Operator

9.14.1. Overview

This example is a containerized AI inference application, developed for use as one of the operators in the Clara Deploy pipelines. The application is built on the base AI application container, which provides the application framework to deploy Clara Train TLT trained models. The same execution configuration file, set of transform functions, and simple inference logic are used; however, inference is performed on the TensorRT Inference Server.

9.14.2. Inputs

This application, in the form of a Docker container, expects an input folder (/input by default), which can be mapped to the host volume when the Docker container is started. This folder must contain a 16-bit PNG image file representing a chest x-ray (CXR).

9.14.3. Outputs

After an image is classified, the operator saves the output as a new image with the classification labels burnt-in on top of the image. The application saves the segmentation results to an output folder (/output by default).

The model supports 15 categories:

  • Nodule

  • Mass

  • Distortion of Pulmonary Architecture

  • Pleural Based Mass

  • Granuloma

  • Fluid in Pleural Space

  • Right Hilar Abnormality

  • Left Hilar Abnormality

  • Major Atelectasis

  • Infiltrate

  • Scarring

  • Pleural Fibrosis

  • Bone/Soft Tissue Lesion

  • Cardiac Abnormality

  • COPD

The top three class categories, with probabilities, are burnt-in to the upper-left corner of the output image. If the probability is high enough (0.5), a category is written out with red color; otherwise, yellow color is used.

The name of each output file has the pattern output-<original file name>.png. The operator also outputs a CSV file (output-<original file name>.csv) that includes the input file path and top three classifications with probabilities (e.g., Granuloma:0.68,Nodule:0.22,COPD:0.02)

9.14.4. AI Model

The application uses the classification_chestxray_v1 model, which uses the tensorflow_graphdef platform. The input tensor is of shape 256x256x3 with a single channel, and the output tensor is of shape 15.

The NVIDIA® Clara Train Transfer Learning Toolkit (TLT) for Medical Imaging provides pre-trained models unique to medical imaging, with additional capabilities such as integration with the AI-assisted Annotation SDK for speeding up annotation of medical images. This allows access to AI-assisted labeling [Reference].

The application uses the classification_chestxray_v1 model provided by the NVIDIA Clara Train TLT for chest x-ray classification, which is converted from the TensorFlow Checkpoint model to tensorflow_graphdef using the TLT model export tool.

You can download the model using the following commands:

# Download NGC Catalog CLI
wget https://ngc.nvidia.com/downloads/ngccli_cat_linux.zip && unzip ngccli_cat_linux.zip && rm ngccli_cat_linux.zip ngc.md5 && chmod u+x ngc

# Configure API key (Refer to https://docs.nvidia.com/ngc/ngc-getting-started-guide/index.html#generating-api-key)
./ngc config set

# Download the model
./ngc registry model download-version nvidia/med/classification_chestxray:1

Note: NGC Catalog CLI is needed to download models without Clara Train SDK: Please follow the NGC documentation to configure the CLI API key.

Detailed model information can be found at (downloaded model folder)/docs/Readme.md.

This application also uses the same transforms library and configuration file for the validation/inference pipeline during TLT model training. The key model attributes (e.g. the model name and network input dimensions), are saved in the config_inference.json file and consumed by the application at runtime.

9.14.4.1. NVIDIA TensorRT Inference Server (TRTIS)

This application performs inference on the NVIDIA TensorRT Inference Server (TRTIS), which provides a cloud inferencing solution optimized for NVIDIA GPUs. The server provides an inference service via an HTTP or gRPC endpoint, allowing remote clients to request inferencing for any model managed by the server.

9.14.5. Directory Structure

The directories in the container are shown below. The application code under /App is from the base container, except for the files in the config directory, which is model specific. The sdk_dist directory contains the Clara Train TLT transforms library. The medical directory contains compiled modules from Clara Train TLT, and the writer directory contains a specialized writer that saves classification results to a PNG image file.

/ai
├── sdk_dist/
└── app_base_inference
    ├── config
    │   ├── config_render.json
    │   ├── config_inference.json
    │   ├── __init__.py
    │   └── model_config.json
    ├── public
    │   └── docs
    │       └── README.md
    ├── writers
    │   ├── __init__.py
    |   ├── classification_result_writer.py
    │   ├── mhd_writer.py
    │   └── writer.py
    ├── app.py
    ├── Dockerfile
    ├── executor.py
    ├── logging_config.json
    ├── main.py
    ├── medical
    └── requirements.txt
/input
/output
/publish

9.14.6. Executing Operator Locally

If you want to see the internals of the container and manually run the application, follow these steps:

  1. Start the container in interactive mode. See the next section on how to run the container, and replace the docker run command with the following:

    docker run --entrypoint /bin/bash
    
  2. Once in the Docker terminal, ensure the current directory is /app.

  3. Execute the following command:

    python ./app_base_inference/main.py
    
  4. When finished, type exit.

9.14.7. Executing Operator in Docker

9.14.7.1. Prerequisites

  1. Ensure the Docker image of TRTIS has been imported into the local Docker repository with the following command:

    docker images
    
  2. Look for the image name TRTIS and the correct tag for the release (e.g. 19.08-py3).

  3. Download both the input dataset and the trained model from the MODEL SCRIPTS section for this container on NGC, following the steps in the Setup section.

9.14.7.2. Step 1

Change to your working directory (e.g. test_chestxray).

9.14.7.3. Step 2

Create, if they do not exist, the following directories under your working directory:

  • input containing the input image file

  • output for the segmentation output

  • publish for publishing data for the Render Server

  • logs for the log files

  • models for the model repository. Copy the contents of the classification_chestxray_v1 folder into this model repository folder.

9.14.7.4. Step 3

In your working directory, create a shell script (e.g. run_chest.sh or another name if you prefer), copy the sample content below, and save it.

#!/bin/bash

# Copyright (c) 2019, NVIDIA CORPORATION.  All rights reserved.
#
# NVIDIA CORPORATION and its licensors retain all intellectual property
# and proprietary rights in and to this software, related documentation
# and any modifications thereto.  Any use, reproduction, disclosure or
# distribution of this software and related documentation without an express
# license agreement from NVIDIA CORPORATION is strictly prohibited.

# Define the name of the app (aka operator); assumed the same as the project folder name
APP_NAME="app_chestxray"

# Define the TenseorRT Inference Server Docker image, which will be used for testing
# Use either local repo or NVIDIA repo
TRTIS_IMAGE="nvcr.io/nvidia/tensorrtserver:19.08-py3"

# Launch the container with the following environment variables
# to provide runtime information
export NVIDIA_CLARA_TRTISURI="localhost:8000"

# Define the model name for use when launching TRTIS with only the specific model
MODEL_NAME="classification_chestxray_v1"

# Create a Docker network so that containers can communicate on this network
NETWORK_NAME="container-demo"

# Create network
docker network create ${NETWORK_NAME}

# Run TRTIS(name: trtis), maping ./models/${MODEL_NAME} to /models/${MODEL_NAME}
# (localhost:8000 will be used)
nvidia-docker run --name trtis --network ${NETWORK_NAME} -d --rm --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 \
    -p 8000:8000 \
    -v $(pwd)/models/${MODEL_NAME}:/models/${MODEL_NAME} ${TRTIS_IMAGE} \
    trtserver --model-store=/models

# Wait until TRTIS is ready
trtis_local_uri=$(docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' trtis)
echo -n "Wait until TRTIS ${trtis_local_uri} is ready..."
while [ $(curl -s ${trtis_local_uri}:8000/api/status | grep -c SERVER_READY) -eq 0 ]; do
    sleep 1
    echo -n "."
done
echo "done"

export NVIDIA_CLARA_TRTISURI="${trtis_local_uri}:8000"

# Run ${APP_NAME} container
# Launch the app container with the following environment variables internally,
# to provide input/output path information
docker run --name ${APP_NAME} --network ${NETWORK_NAME} -it --rm \
    -v $(pwd)/input:/input \
    -v $(pwd)/output:/output \
    -v $(pwd)/logs:/logs \
    -v $(pwd)/publish:/publish \
    -e NVIDIA_CLARA_TRTISURI \
    -e NVIDIA_CLARA_NOSYNCLOCK=TRUE \
    ${APP_NAME}

echo "${APP_NAME} is done."

# Stop TRTIS container
echo "Stopping TRTIS"
docker stop trtis > /dev/null

# Remove network
docker network remove ${NETWORK_NAME} > /dev/null

9.14.7.5. Step 4

Execute the script as shown below and wait for the application container to finish:

./run_chest.sh

9.14.7.6. Step 5

Check for the following output files:

Classification results in the output directory:

  • output-<input file name>.csv

  • output-<input file name>.png

9.14.7.7. Step 6

To visualize the segmentation results, or the rendering on Clara Dashboard, please refer to sections on visualization.