LesionFinder

Development of Pigmented Skin Lesions Segmentation System using thresholds and Morphological Operations

Abstract

The project aims to develop a mole and melanoma detection system using Matlab, evaluating segmentation techniques to identify skin lesions. The system focuses on simple and explainable techniques that require low computational resources, such as thresholding and morphological segmentation, to be useful both in clinical practice and in the education of biomedical engineering students. Computational algorithms and image processing techniques can detect abnormalities, including skin lesions. Research in computational methods to analyze skin abnormalities is growing, but current diagnostic accuracy remains insufficient. The methodology includes needs assessment, architecture design, software development, testing and validation. Images from the HAM10000 dataset were used for testing, and the system was validated using metrics such as sensitivity, specificity, and precision, comparing the results with the ImageJ software. The system showed high precision in detecting benign nevi but difficulties with melanomas, suggesting that morphological techniques are insufficient for clinical detection.