2021/09/15 by Sivaramakrishnan Sankarapandian, Sankarapandian, Sivaramakrishnan, Saul A. Kohn +21 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · Medicine · #AI in cancer detection #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #Cutaneous Melanoma Detection and Management #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2109.07554
openalex publication_date 2021/09/15 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Although melanoma occurs more rarely than several other skin cancers,\npatients' long term survival rate is extremely low if the diagnosis is missed.\nDiagnosis is complicated by a high discordance rate among pathologists when\ndistinguishing between melanoma and benign melanocytic lesions. A tool that\nallows pathology labs to sort and prioritize melanoma cases in their workflow\ncould improve turnaround time by prioritizing challenging cases and routing\nthem directly to the appropriate subspecialist. We present a pathology deep\nlearning system (PDLS) that performs hierarchical classification of digitized\nwhole slide image (WSI) specimens into six classes defined by their\nmorphological characteristics, including classification of "Melanocytic\nSuspect" specimens likely representing melanoma or severe dysplastic nevi. We\ntrained the system on 7,685 images from a single lab (the reference lab),\nincluding the the largest set of triple-concordant melanocytic specimens\ncompiled to date, and tested the system on 5,099 images from two distinct\nvalidation labs. We achieved Area Underneath the ROC Curve (AUC) values of 0.93\nclassifying Melanocytic Suspect specimens on the reference lab, 0.95 on the\nfirst validation lab, and 0.82 on the second validation lab. We demonstrate\nthat the PDLS is capable of automatically sorting and triaging skin specimens\nwith high sensitivity to Melanocytic Suspect cases and that a pathologist would\nonly need between 30% and 60% of the caseload to address all melanoma\nspecimens.\n