Comparison of artificial vision algorithms in the classification of skin diseases using neural networks

Abstract

This study evaluates three tools for the detection of various dermatological diseases, including melanomas, chickenpox, measles, lupus, herpes, scabies, and monkeypox. The results indicate that Orange Data Mining consistently demonstrated high precision (98.2% in training, 99.8% in validation), while Azure Custom Vision achieved moderate precision (88.6% in training, 62.7% in validation), and the CNN showed lower precision (28.14% in validation). The objective of the research is to provide a diagnostic support tool for medical personnel in Level 1 clinics in Mexico, helping them detect sick patients and channel them to Level 2 clinics or specialists, ultimately leading to an improvement in the quality of life for patients who do not have the resources or a Level 2 hospital near their residence.