Medical Image Classification Deep Learning Pipeline

Construct an end-to-end medical image classification system using TensorFlow and Keras that can diagnose potential pathologies from radiological images. The system must include data augmentation, transfer learning from pre-trained medical imaging models, and a robust validation framework. Implement multi-class classification with interpretable results, including confidence intervals and potential diagnostic recommendations. Include mechanisms for handling imbalanced medical datasets and generating DICOM-compatible output.