CNN-GAN Cancer Image Classifier

I have four independent, pre-labelled imaging datasets—MRI, ultrasound, mammogram and histology—each split into the two diagnostic classes that matter most to clinicians: benign and malignant. My goal is to write a single, reusable codebase that I can launch against any one of these datasets and obtain the highest possible classification accuracy.

The pipeline must combine a convolutional backbone for feature extraction, a GAN-based data-augmentation module to balance classes and enlarge the training set, and Grad-CAM to visualise the discriminative regions that drive each prediction. Because the datasets differ in resolution and appearance, the code should accept modality-specific preprocessing parameters but keep the core architecture, training loop and evaluation logic identical; I will simply point the script to the chosen dataset and let it run.

Key expectations
• End-to-end Python code (TensorFlow or PyTorch) that trains, validates and tests the model, outputs accuracy, AUC and confusion matrix, and saves Grad-CAM heat-maps alongside original images.
• Clear configuration section so I can switch between the four datasets without touching the model code itself.
• Well-commented notebook or script plus a short README explaining environment setup and how to reproduce your results.

Deliverables will be considered complete when the code executes successfully on at least one sample from each modality and the Grad-CAM overlay correctly highlights tumour regions.

If this challenge matches your skill set in CNNs, GANs and medical-image explainability, let’s collaborate.