Algorithms for Pigmented Lesion Screening and Detection

The study addresses the need for better physician training by providing specific accurate clinical descriptions of pigmented lesions including malignant melanoma. In Phase I, we will develop clinical algorithms to precisely describe malignant melanoma. In this phase, our aims include 1) location of early melanomas within the melanoma color band, definition of variegation, precise color description and quantization, 2) automatic border detection using global image features and automatic induction, 3) seeking optimal measures for analysis of asymmetry and irregularity and apply to pigmented lesion images, and 4) showing the importance of these features. Phase I will demonstrate what are the relative weights of critical features to be combined to generate a simple clinical rule for melanoma diagnosis. Phase II will add one more analytic feature: computer-determined visual texture. What texture measures best help to distinguish melanoma from irregular seborrheic keratoses? Analysis as in Phase I will be performed for benign pigmented lesions and will include development of software for computer-assisted instruction for improved clinical diagnosis of pigmented lesions. Future application of computer vision diagnostic assistance systems might enable low-cost screening of large groups for pigmented lesions. These could include routine hospital admissions, nursing home patients at annual intervals and public skin cancer screenings.