Design and Optimisation of a CNN-Based Chest X-Ray Classific…

This assignment focuses on the MedScan AI scenario: classifying Chest X-ray images into 3 classes (Normal, Pneumonia, COVID-19) using a CNN/transfer learning approach. Task 1 requires designing and explaining a 5-layer CNN, and justifying Softmax for 3-class output plus whether to use Transfer Learning (e.g., ResNet50) due to limited data (~5,000 images). Task 2 analyzes training where validation loss rises after ~epoch 10 (overfitting), proposes 3 regularization methods (e.g., Dropout, Data Augmentation, Early Stopping), and states the 3-class cross-entropy loss: I will also discuss suitable evaluation metrics (Precision/Recall/F1) and ethical considerations.