Deep Learning for Skin Lesion Segmentation

Novel pre-processing and deep learning techniques for melanoma detection

Project Supervisor: Professor Jagath Rajapakse (Nanyang Technological University)

Developed an advanced pipeline for skin lesion segmentation combining novel image pre-processing, modified deep learning architecture, and post-processing techniques. The approach demonstrated 7% improvement over baseline methods and achieved state-of-the-art segmentation accuracy for identifying cancerous regions.

Pipeline Components:

  1. Pre-processing: Custom image enhancement to improve lesion visibility
  2. Deep Learning: Modified CNN architecture (Keras/TensorFlow backend)
  3. Post-processing: Morphological operations to refine segmentation boundaries

Results: Significant improvement in melanoma detection accuracy, with practical applications in dermatological screening.

Code & Resources: