Application of artificial intelligence as a tool for optimization in the diagnosis of melanoma

Abstract

INTRODUCTION: Melanoma, a skin cancer with increasing global incidence, is prevalent in regions with light-skinned populations. Its ability to metastasize and high mortality rate highlight the need for early detection, for effective intervention and potential reduction of risk of metastasis Strategies such as self-examinations and dermatological check-ups are essential. Artificial intelligence (AI) has emerged as a tool with greater precision and speed in the analysis of skin images.

OBJECTIVES: 1. Achieve optimization in the early diagnosis of melanoma through AI as a collaborative tool in dermatoscopy. 2. Determine the diagnostic accuracy of Artificial Intelligence as a tool in dermatoscopy vs. the diagnosis of the human eye through a specialized dermatologist.

MATERIAL AND METHODS: This is a prospective, cross-sectional cohort study.

At Avedian we developed a deep learning model for the detection of melanoma. We trained the model with 12,000 images (sample size n: 12,000 images; the required sample size according to Machin' s calculation is 526 images) obtained from the database obtained from the international skin imagine collaboration, and we carried out a comparative study of image classification of melanomas and healthy moles. On the other hand, we gave 40 images to 2 dermatologists specialized in the detection of nevi and melanomas, who were asked to classify these images according to their experience and clinical criteria, in binary form.

RESULTS: The comparative study showed an overall accuracy of the AI ​​deep learning model of 82.5%, with 85% for nevi and 80% for melanomas, compared to the binary classification system of dermatologists,

(61%), with a sensitivity of 85% and specificity of 82%, confidence level of 95%.

CONCLUSIONS: The integration of AI in early detection promises significant advances in the diagnosis and treatment of melanoma and allows for the optimization of early diagnosis tools.