Medical Image Processing with Deep Learning

Abstract

This thesis focuses on melanoma detection from dermoscopic images using deep learning for early diagnosis. EfficientNet-B3 was used in this research work with transfer learning by initializing the pre-trained weights on ImageNet for better feature extraction. Data augmentation, Early Stopping and K-fold cross-validation techniques were employed to avoid overfitting. Ensemble learning was used to achieve better accuracy and evaluate the overall model. The model has achieved an accuracy of 88.96%.