Medical Imaging · Deep Learning
Improved Pneumothorax Classification using DCGAN Augmentation
A DCGAN synthesizes pneumothorax chest X-rays to balance a scarce SIIM dataset, and a ResNet-18 classifier is fine-tuned on the augmented data, reaching 98.6% accuracy and a perfect ROC-AUC. Grad-CAM and t-SNE confirm the model learned real clinical features.
- 98.60% Train Accuracy
- 1.000 ROC AUC
Stack PyTorch / ResNet-18 / DCGAN / Grad-CAM
Problem
Pneumothorax, a collapsed lung, is diagnosable from chest radiographs but difficult to read even for experienced radiologists, and labeled positive cases are scarce relative to negatives. That class imbalance, combined with the barriers to sharing clinical data, makes it hard to train a classifier that generalizes rather than memorizing the majority class.
Approach
Rather than rely on rotations and rescaling, which are too simple to capture real clinical variation, the project addresses scarcity by synthesizing new positive examples. A deep convolutional GAN (DCGAN) is trained on the SIIM-ACR data to generate realistic synthetic pneumothorax X-rays, which are folded back into the training set to balance the positive and negative classes. A pretrained ResNet-18 classifier is then fine-tuned on the augmented data.
To validate the model, Grad-CAM confirms that predictions are driven by plausible thoracic regions, and t-SNE shows the learned feature space separates the classes cleanly.
Results
- 98.60% training accuracy
- ROC-AUC = 1.000
- Grad-CAM saliency and t-SNE embeddings confirming the model learned clinically meaningful features
The result demonstrates that DCGAN-based augmentation is an effective approach for medical imaging tasks where labeled data is the primary constraint.