Making AI Work in Mohs Micrographic Surgery Workflow

A poster session at the 2023 American College of Mohs Surgery Meeting in Seattle, Washington showed the results of a retrospective cohort study with promising results in the integration of artificial intelligence (AI) in Mohs Micrograpic surgery (MMS) for basal cell carcinoma (BCC).

While no patient case is the same, Mohs surgeons typically use the same workflow for each case: remove the tumor, allow the histotechnologist perform cryofreezing, sectioning, staining, and conclude with a histologic analysis. The current, long-time workflow allows Mohs surgeons to observe 100% of affected skin margins in real-time and prevent cancer-positive margins during post-operative analysis. There are a few downsides to this process including the availability of pathologists with cancer-specific experience, a surgeon’s knowledge of tumor mapping, and availability of histotechnicians for crucial tasks such as grossing, inking and, sectioning tissue. To overcome these obstacles in the MMS workflow, study researchers wanted to know if an AI platform could facilitate real-time histologic analysis and tumor mapping.

carrascosaoscar/Envato Elements

The research took place at a single site MMS clinic where Arctic AI software framework was developed. The framework was created by building 3-D tissue models and specimen measurements with smartphone videos and a turntable platform to analyze tissue specimens (n=17). Tumors were also localized with 1065 tissue sections, 32,763 cell annotations were identified, and 381 tissue sections were analyzed for tissue completeness. The tissue then had to be oriented and mapped by using detected surgical inking patters to morph Whole Slide Imaging (WSI) histologic tissue sections and tissue orientation into the shape of the MMS map.

"3-D tissue reconstruction generated accurate and complete 3-D renderings with precise measurements,” study researchers reported. “Tumor localization obtained an area under the curve (AUC) measurement of .97 on a test set of 121 held out slides, nuclei detection and classification localized cells with Dice=0.86, and analysis of tissue completeness obtained an AUC of 0.84.”

The AI platform also showed a correct tissue orientation 95% of the time. The 5% error rate was caused by insufficient inking of the specimens, and tissue mapping achieved an accuracy score of 99.2%. A total of 41 cases and 121 slides were observed during the assessment of the Arctic AI platform and had an average execution time of 72 seconds per slide obseration and 78 seconds total per case (95% CI[66-88]).

In conclusion, the promising results used in the removal of BCC will still need to show significant improvement in efficacy compared to current workflow systems in order to get funded by investors and stakeholders to be utilized broadly in Mohs surgery.

Earlier today, Dermatology Times® posted a poll and wants to know whether you think the use of AI in Mohs surgery can enhance workflow. We would love to hear from you!

Reference

1. Davis M, Levy J, Chacko R, et al. A deep learning algorithm for integration of artificial intelligence in the Mohs Micrographic surgery workflow for treatment of basal cell carcinoma. Poster presented at the 2023 American College of Mohs Surgery Meeting;May 4-7, 2023; Seattle, WA. https://acms.planion.com/Web.User/AbstractDet?ACCOUNT=ACMS&ABSID=117897&CONF=AM2023&ssoOverride=OFF&CKEY=8A7Y39186