The original research article, High-fidelity detection, subtyping, and localization of five skin neoplasms using supervised and semi-supervised learning, describes a novel computational pathology algorithm (Mihm) that detects and classifies both potentially malignant and benign lesions from whole slide images with extremely high sensitivity. In addition, this high-fidelity AI model subtypes lesions, localizes lesions on the whole slide image, and informs on margin status based on proximity of the lesion to the non-epidermal edge. Mihm is the cornerstone of an end-to-end dermatopathology tool that comprises a dedicated lab for standardized tissue processing, dermatopathologists for expert annotations and sample acquisition, and back- and front-end software, the latter of which includes an interactive user interface that displays predictions of algorithmic outputs and regions of interest. Mihm was trained on a large-scale training dataset that seamlessly combines both supervised and semi-supervised learning.
Technical & Academic Writing, Illustration Development, Scientific Editing
Journal of Pathology Informatics