Identification and Estimation of Clinical Indices Useful for the Diagnosis of Melanoma from Macroscopic Images

Melanoma is an extremely aggressive form of skin cancer. When not promptly detected and treated, it can quickly metastasize, leading to unfavourable prognostic outcomes. Achieving early melanoma diagnosis relies heavily on accurate and thorough skin analysis, made by an expert dermatologist. To address subjective judgments and time-expensive exams, a novel screening and diagnostic method utilising photogrammetry-derived images of skin lesions has been devised. This innovative approach is based on the acquisition of macroscopic images, depicting a large portion of the patient body, and enables the creation of a three-dimensional model of the patient, allowing for the extraction of corresponding images of each individual lesion. This thesis aims to quantitatively assess the asymmetry, the irregularity of the border and the color of skin lesions through the analysis of segmented macroscopic images, contributing to the development of an automated diagnostic tool useful to the clinician for melanoma identification. The analysis was conducted on a dataset comprising images of healthy skin lesions and lesions reported as suspicious by dermatologists among which nine cases were confirmed as melanomas by biopsy. By utilizing algorithms to objectively compute asymmetry and border irregularity parameters, coupled with an in-depth analysis of color features associated with melanocytic lesions, the investigation unveiled statistically significant differences in these attributes between benign and suspicious lesions. Indeed, statistical tests confirmed distinctive distributions of these parameters between the two skin lesion populations. These findings underscore the potential of automated diagnostic tools derived from macroscopic images in effectively identifying suspicious lesions, thus contributing to early melanoma detection strategies.