Leveraging AI to Assist Cancer Diagnosis

Basal cell carcinoma is the most common skin cancer worldwide and continues to increase in incidence [1]. Surgical excision is a common modality of treatment and the frozen section can be judiciously used intraoperatively to determine the margin status of the excision, particularly for the lesions in the critical anatomical sites such as the nose, cheeks, eyelids, chin, lips, and forehead [2]. For these locations, a persistent/recurrent disease from an incomplete primary excision would demand a far more aggressive and disabling secondary surgery [2].

Margin assessment of basal cell carcinoma using the frozen section is a common task of pathology intraoperative consultation. Although frequently straightforward, the determination of the presence or absence of basal cell carcinoma on the tissue sections can sometimes be challenging.

Collaborating with Carolinas Dermatology and Plastic Surgery and Beaumont Health, we explored if a deep learning model trained on mobile phone-acquired frozen section images can have adequate performance for future deployment. Our physicians use smartphones to capture thousands of images of frozen sections performed for basal cell carcinoma margin status. The photos were taken at 100x magnification (10x objective).

The images were downscaled from a 4032 x 3024 pixel resolution to 576 x 432 pixel resolution. Semantic segmentation algorithm Deeplab V3 with Xception backbone was used for model training. The model uses an image as input and produces a 2-dimensional black and white output of prediction of the same dimension; the areas determined to be basal cell carcinoma were displayed with white color, on a black background. Any output with the number of white pixels exceeding 0.5% of the total number of pixels is deemed positive for basal cell carcinoma. On the test set, the model achieves an area under curve of 0.99 for the receiver operator curve and 0.97 for precision-recall curve at the pixel level. The accuracy of classification at the slide level is 96%.

The deep learning model trained with mobile phone images shows satisfactory performance characteristics, and thus demonstrates the potential for deploying as a mobile phone app to assist in frozen section interpretation in real-time.