Abstract
42753 Artificial Intelligence Automates Detection of Basal Cell Carcinoma in Frozen Section Biopsies
Journal of the American Academy of Dermatology, Vol.89(3 Suppl.), pp.AB17-AB17
09/2023
DOI: 10.1016/j.jaad.2023.07.074
Abstract
The evolution of digital pathology has created opportunities to build artificial intelligence (AI)-powered systems to provide real-time diagnostic feedback for clinicians1. Several studies have presented deep learning models for the automated detection and segmentation of basal cell carcinoma (BCC) on digitized Mohs frozen sections2-4. We present a deep learning model that was trained on digital biopsy frozen section whole slide images (WSIs) selected at random regardless of tissue quality or completeness. Two hundred WSIs, of which 88 (44%) contained BCC, from 1769 patient encounters at Mayo Clinic Rochester from January 2019 to March 2020 were included for algorithm training and testing. Structures of interest (e.g., BCC, actinic keratosis, hair follicle, squamous cell carcinoma, etc.) were annotated by a Mohs fellow, senior dermatopathologist, and senior Mohs surgeon. Annotations were split into a 70-15-15 train-tune-test sets and divided into 896x896-pixel patches to train a deep neural network (DNN) pretrained on ImageNet5. Patch-level predictions from the DNN were used to fit a logistic regression (LR) model to make WSI-level predictions on the presence of BCC. On the final test set, the model sensitivity was 0.73 and specificity was 0.88. The area under the curve (AUC) for the receiver-operator curve (ROC) was 0.9. Our findings suggest the feasibility of algorithmic tools for detection of BCC in biopsy frozen sections despite variation in quality and tumor type. This may eventually provide preoperative clinical decision support to enhance the workflow and patient-provider experience in Mohs micrographic surgery.
Details
- Title: Subtitle
- 42753 Artificial Intelligence Automates Detection of Basal Cell Carcinoma in Frozen Section Biopsies
- Creators
- Yong-hun Kim - Mayo Clinic Alix School of MedicineDennis Murphree - Mayo Clinic in ArizonaKirk Sidey - University of IowaNneka Comfere - Mayo ClinicNahid Vidal - Mayo Clinic in Arizona
- Resource Type
- Abstract
- Publication Details
- Journal of the American Academy of Dermatology, Vol.89(3 Suppl.), pp.AB17-AB17
- DOI
- 10.1016/j.jaad.2023.07.074
- ISSN
- 0190-9622
- eISSN
- 1097-6787
- Publisher
- Elsevier Inc
- Language
- English
- Date published
- 09/2023
- Academic Unit
- Dermatology
- Record Identifier
- 9984466774202771
Metrics
12 Record Views