Abstract

Glass pathology slides are increasingly being scanned using high-resolution scanners which allow for assessment and interpretation over a computer interface, as opposed to oculars on a microscope. The COVID pandemic increased this practice, and the expectation is that it will only become more widespread. As slide scanning and electronic interpretation becomes more widespread, there is new opportunity for artificial image interpretation. We sought to apply this technology for detection of sentinel node metastases, very important to the patient and treating clinicians but tedious and time-consuming for the pathologist. Our proof of principle study involved a training set of 32 slides stained for SOX10 including 16 positives and 16 negatives which were scanned and evaluated through image analysis. SOX10 immunohistochemistry was chosen because it tends to be highly sensitive and crisp, with minimal background. A test set of eight slides (four positive slides and 4 negative slides) was then evaluated. On this small set we were able to achieve a separation accuracy of 87.5%. All negative slides were interpreted correctly as negative. One positive slide, with very minimal melanoma metastatic burden consisting of only a few cells was interpreted incorrectly as negative. Our group is currently in the process of expanding this proof of principal pilot to include significantly more slides, both in the training group and in the test group. We believe that this endeavor may be beneficial to pathologists in the future to detect melanoma sentinel lymph node metastases in a timely and more efficient manner.