Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reproducing and Improving CheXNet: Deep Learning for Chest X-ray Disease Classification

Daniel J. Strick, Carlos Garcia, Anthony Huang, Thomas Gardos

TL;DR

This work reproduced an algorithm known as CheXNet, as well as explored other algorithms that outperform CheXNet's baseline metrics, and evaluated model performance using the F1 score and AUC-ROC, both of which are critical metrics for imbalanced, multi-label classification tasks in medical imaging.

Abstract

Deep learning for radiologic image analysis is a rapidly growing field in biomedical research and is likely to become a standard practice in modern medicine. On the publicly available NIH ChestX-ray14 dataset, containing X-ray images that are classified by the presence or absence of 14 different diseases, we reproduced an algorithm known as CheXNet, as well as explored other algorithms that outperform CheXNet's baseline metrics. Model performance was primarily evaluated using the F1 score and AUC-ROC, both of which are critical metrics for imbalanced, multi-label classification tasks in medical imaging. The best model achieved an average AUC-ROC score of 0.85 and an average F1 score of 0.39 across all 14 disease classifications present in the dataset.

Reproducing and Improving CheXNet: Deep Learning for Chest X-ray Disease Classification

TL;DR

This work reproduced an algorithm known as CheXNet, as well as explored other algorithms that outperform CheXNet's baseline metrics, and evaluated model performance using the F1 score and AUC-ROC, both of which are critical metrics for imbalanced, multi-label classification tasks in medical imaging.

Abstract

Deep learning for radiologic image analysis is a rapidly growing field in biomedical research and is likely to become a standard practice in modern medicine. On the publicly available NIH ChestX-ray14 dataset, containing X-ray images that are classified by the presence or absence of 14 different diseases, we reproduced an algorithm known as CheXNet, as well as explored other algorithms that outperform CheXNet's baseline metrics. Model performance was primarily evaluated using the F1 score and AUC-ROC, both of which are critical metrics for imbalanced, multi-label classification tasks in medical imaging. The best model achieved an average AUC-ROC score of 0.85 and an average F1 score of 0.39 across all 14 disease classifications present in the dataset.
Paper Structure (13 sections, 4 figures, 4 tables)

This paper contains 13 sections, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Approach
  4. Datasets
  5. Evaluation Results
  6. Conclusion
  7. Role of the Funding Source
  8. Generative AI Usage Statement

Figures (4)

  • Figure 1: (Left) An example of an X-Ray from the NIH ChestX-ray14 with Hernia and Infiltration as its ground truth findings. (Center) Grad-CAM visualizations of the same X-Ray image. (Right) DACNet’s predictions of the 5 most likely findings in the image.
  • Figure 2: Graphs from wandb.ai showing comparisons of the three final models: DACNet.py, vit_transformer.py, and replicate_chexnet.py. The average validation AUC score across all diseases is shown across the entire training run.
  • Figure 3: Graphs from wandb.ai showing comparisons of the three final models: DACNet.py, vit_transformer.py, and replicate_chexnet.py. The average validation F1 score across all diseases is shown across the entire training run.
  • Figure 4: Graphs from wandb.ai showing comparisons of the three final models: DACNet.py, vit_transformer.py, and replicate_chexnet.py. The average validation loss across all diseases is shown across the entire training run.