Computer-Aided Pigment Network Detection in Dermoscopic Images for Melanoma Screening

Abstract

Early detection of melanoma, the deadliest form of skin cancer, is vital for successful treatment. A key diagnostic indicator is the pigment network, yet its accurate identification in dermoscopic images is often hindered by noise and obstructive hair artifacts. This thesis addresses these challenges by introducing a novel image processing framework for computer-aided pigment network detection. Our approach begins with advanced filtering techniques to suppress noise and remove hair, significantly enhancing image clarity. The preprocessed images are then converted into binary formats to isolate critical structures. Finally, the system detects the pigment network and extracts key quantitative features such as diameter and radius—to assess melanoma risk. The proposed method achieves state-of-the-art performance with an overall accuracy of 85.5%, an average precision of 0.89, and a recall of 0.87, demonstrating its potential as an efficient and reliable tool for automated diagnosis.