Cellular pattern recognition towards discrimination of normal skin from melanoma in non-invasive confocal imaging

Cellular histopathological melanoma screening is critical but expensive/invasive. Confocal screening is cheap/noninvasive but data interpretation remains difficult. Human terminology for biological features is insufficient to fully exploit the diagnostic value, so we propose automated quantitative morphometry. Normal diagnostic traits include a regularly organized spinous keratinocyte matrix on an underlying smooth basal keritinocyte layer. Computational identification of dark nuclei in spinous keratinocytes and bright pigmented basal keratinocytes yields two distinct regions: basal and super-basal. These independent algorithms usually yield complementary regions but occasionally overlap or leave gaps. Improved microanatomical discrimination will yield a better diagnostic map to evaluate morphology for cancer detection.