clara_pt_chest_xray_classification

Model Overview

A pre-trained densenet121 model for disease pattern detection in chest X-rays.

Note: The 4.1 version of this model is only compatible with the 4.1 version of the Clara Train SDK container

Model Architecture

The model is trained using a densenet121 model [1] for disease pattern detection in chest X-rays [2,3].

The overall pipeline of the trained model is shown below:

Training

The training was performed with the following:

• Script: train.sh
• GPU: At least 16GB of GPU memory.
• Actual Model Input: 256 x 256
• AMP: True
• Optimizer: Adam
• Learning Rate: 1e-4
• Loss: Binary Cross Entropy (BCE) Loss

Dataset

This model is trained using PLCO training data and evaluated on the PLCO validation data.

• Target: 15 chest disease classes
• Task: Classification
• Modality: X-ray
• Size: 179,589 images (139,509 Training, 18,000 Validation, 22,080 Testing)

You can apply for access to the dataset at: https://biometry.nci.nih.gov/cdas/learn/plco/images/

Data Preparation

The data must be converted to 16-bit png images before training.

Performance

This model achieves an Averaged AUC over all disease categories on the testing set of 0.832888923438338 (reproduce the results by changing VAL_DATALIST_KEY in environment.json to 'testing').

Training

A Graph showing the training accuracy over 40 Epochs.

Validation

A graph showing the validation accuracy over 40 Epochs.

How to Use this Model

The model was validated with NVIDIA hardware and software. For hardware, the model can run on any NVIDIA GPU with memory greater than 16 GB. For software, this model is usable only as part of Transfer Learning & Annotation Tools in Clara Train SDK container. Find out more about Clara Train at the Clara Train Collections on NGC.

Full instructions for the training and validation workflow can be found in our documentation.

Input

Input: 16-bit CXR png with a shape of 256 x 256

Preprocessing:

1. Randomly cropping (random center and size) with a ROI size 230 x 230 as output
2. Resize images to 256 x 256
3. Randomly rotation with a maximum of 7 degrees
4. Normalizing to unit std with zero mean

Output

Output: 15 binary labels

Each bit is corresponding to the prediction of: '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', and 'COPD'.

Limitations

This training and inference pipeline was developed by NVIDIA. It is based on a classification model developed by NVIDIA researchers. This research use only software has not been cleared or approved by FDA or any regulatory agency. Clara pre-trained models are for developmental purposes only and cannot be used directly for clinical procedures.

References

[1] Huang, Gao, et al. "Densely connected convolutional networks." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017. https://arxiv.org/abs/1608.06993.

[2] Wang, Xiaosong, et al. "Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017. https://arxiv.org/abs/1705.02315.

[3] Chen, Haomin, et al. "Deep hierarchical multi-label classification of chest X-ray images." In International Conference on Medical Imaging with Deep Learning, pp. 109-120. PMLR, 2019. https://proceedings.mlr.press/v102/chen19a.html

License

End User License Agreement is included with the product. Licenses are also available along with the model application zip file. By pulling and using the Clara Train SDK container and downloading models, you accept the terms and conditions of these licenses.