Introduction: Melanoma develops either de novo or from non-obligate precursor lesions known as atypical/dysplastic nevi. Assessment of change in number and morphology of pigmented cutaneous lesions over time is critical to early detection of skin cancers and may provide preliminary signals of efficacy in early phase therapeutic prevention trials for melanoma. Despite the use of total body digital photography for at least 20 years to document the presence of these lesions, as well as recent progress in computer-aided diagnosis of lesions in clinical images, automated methods to characterize the evolution of skin lesions are still lacking. The purpose of this study was to develop and validate a computer vision approach to facilitate detection and quantification of changes in nevi in serial digital photographs. Methods: The ‘DermaViz' mathematical algorithms were developed to register nevi between sequential images and to align images for improved comparison. The technique is based on the bispectrum algorithm, modified to adapt for human skin changes. Adaptive normalization techniques adjust for lighting and skin tone variations. Warping and shear of skin are accommodated by a hierarchical iteration of these algorithms coupled with probabilistic matching techniques for accurate alignment. The technology allows both for improved qualitative comparison by clinicians when the aligned images are toggled between dates, and for digital quantification of changes in (a)symmetry, (b)order, (c)olor, and (d)iameter of the lesions. In this pilot study, serial posterior truncal photographs from 17 patients with multiple atypical nevi and a history of melanoma were obtained from a pre-existing image and nevus biobanking protocol database at our institution. De-identified images were processed and analyzed with DermaViz software, and results were validated by a panel of Melanoma Program clinicians. Results: DermaViz software had a high sensitivity for detection of cutaneous lesions as small as 2mm, which was limited by the quality of the archival photographs. The software registered specific nevi accurately in most cases, with sight errors in a small number of lesions that were primarily located at the edges of the images. In the 17-patient pilot study, registration and alignment of serial images enabled clinicians to identify new and enlarged nevi in 3 to 11 additional patients vs the unregistered images. Quantification with DermaViz correlated with physician assessment of new and enlarged nevi in 90% of evaluated lesions. Conclusion: Software has been designed, applied, and validated that dramatically improves detection of changes in nevi over time and enables quantification of these changes. It helped clinicians to identify numerous changes that were missed in the original unregistered images. We plan to incorporate an expanded ruler and color balance tape in future photographs for improved analyses of border, color, and size changes. Dermaviz will be used in a planned Phase II trial of sulforaphane for therapeutic prevention of melanoma (EA6201).

Citation Format: William F. Maguire, Paul H. Haley, Catherine M. Dietz, Mike Hoffelder, Clara S. Brandt, Robin Joyce, Melissa D. Wilson, Darcy Ploucha, Christopher Minnier, Cindy Sander, Hong Wang, Hassane M. Zarour, Kevin J. Mitchell, Ellen K. Hughes, John M. Kirkwood. Automated image registration and alignment facilitates assessment of change in pigmented lesions of patients at risk for melanoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr LB227.