This proposal will help to improve the accuracy of pathologists diagnosing melanoma and melanocytic lesions. The incidence of melanoma is rising faster than any other cancer, and ~1 in 50 U.S. adults will be diagnosed with melanoma this year alone. Based on previous work, our research team has noted substantial and frequent diagnostic errors in interpreting melanocytic lesions; pathologists disagree in up to 60% of cases of invasive melanoma, which can lead to substantial patient harm due to misdiagnosis. Our proposed work leverages information naturally embedded in digitized slides (whole-slide digital images of glass slides) to use computer technology to improve the diagnosis of melanocytic lesions. Using data from a currently funded NIH study that has documented substantial and concerning diagnostic errors from pathologists, we will digitize and study a well-characterized set of 240 skin biopsy cases that includes a spectrum of benign to invasive melanoma diagnoses. Each biopsy case has been independently interpreted by many practicing U.S. pathologists and also a panel of international experts in dermatopathology, providing a uniquely rich clinical database that is the largest of its kind. In this project, novel computational techniques will be used to analyze digitized slides for the purpose of assisting in the pathologic diagnosis of melanoma and related skin lesions. These techniques include the detection of both cellular-level and architectural features for use in feature-based classification, and exploration of deep neural networks that operate on raw pixel data for the difficult task of mitosis detection. In addition, a machine learning approach will be applied to the digitized slides to determine the histopathological characteristics associated with human diagnostic errors.
Our specific aims are as follows: 1. To design and implement image analysis algorithms to detect clinically important features in digitized slide images of melanocytic skin lesions. 2. To develop classification systems that can categorize digitized slide images into one of five possible diagnostic classes: benign; atypical lesions; melanoma in situ; invasive melanoma stage T1a; and invasive melanoma stage ?T1b. 3. To investigate the characteristics of digitized slide images associated with diagnostic errors by human pathologists using data from both expert and community pathologists. In our proposed study, we are innovatively merging data on how pathologists review and diagnose slides in a clinical setting with image analysis algorithms. This technology has the potential to improve diagnostic accuracy of pathologists by providing an analytical, undeviating review to assist humans in this difficult task.
This study will use computer image analysis techniques to improve our understanding of the causes of diagnostic errors during the interpretation of skin biopsy specimens, as well as seek ways to reduce such errors. As skin biopsies are one of the most common medical procedures performed in the U.S., the results of this study have important implications for patients as these tests are frequently used to guide important treatment recommendations for melanoma and surveillance recommendations for dysplastic nevi.
| Elmore, Joann G; Elder, David E; Barnhill, Raymond L et al. (2018) Concordance and Reproducibility of Melanoma Staging According to the 7th vs 8th Edition of the AJCC Cancer Staging Manual. JAMA Netw Open 1: |
