Medical Image Analysis Pipeline for Early Detection

Design a scalable medical image analysis pipeline capable of detecting early-stage anomalies across multiple imaging modalities (X-ray, MRI, CT scans). Develop a transfer learning approach that can leverage pre-trained computer vision models while adapting to specialized medical imaging datasets. The solution must include robust preprocessing techniques, model validation protocols, and a framework for handling class imbalance in rare disease detection. Demonstrate performance metrics including sensitivity, specificity, and area under the ROC curve.