Identification of Metastatic Primary Cutaneous Squamous Cell Carcinoma Utilizing Artificial Intelligence Analysis of Whole Slide Images

Abstract

Listen

Translate article

Background: Cutaneous squamous cell carcinoma (cSCC) possesses metastatic potential and causes significant mortality. However, clinical assessment of metastasis risk is challenging. We approached this challenge by harnessing artificial intelligence (AI) algorithm to identify metastatic primary cSCCs. Methods: Residual neural network-architectures were trained and fine-tuned to identify metastatic cSCCs on tumor tiles extracted from clinician annotated, hematoxylin and eosin-stained whole slide images representing metastatic (n=45) and non-metastatic (n=59) primary cSCCs. Findings: Several models were trained and tested with best area under the receiver operating characteristic curve (AUROC) of 0·689 when all primary tumors (n=104) regardless of time to metastasis were included. Metastatic primary tumors were further divided into two groups that metastasize rapidly (≤180d) (n=22) or slowly (>180d) after primary tumor detection (n=23). The algorithm was able to predict whether primary tumor was non-metastatic or classified as rapidly metastatic with slide-level AUROC of 0·747. This prediction was superior to blind assessment by pathologist and did not appear to rely on classical clinicopathological features or staging systems. Furthermore, risk factor (RF) model taking into account prediction by AI, Clark’s level and tumor diameter provided higher AUROC (0·917) than other tested RF models and predicted high 5-year disease specific survival (DSS) for patients with cSCC with 0 or 1 RFs (100% and 95·7%) and poor 5-year DSS for patients with cSCCs with 2 or 3 RFs (41·7% and 40·0%). Interpretation: These results provide evidence, that AI recognizes unknown features associated with metastasis and provides added value to challenging clinical risk assessment of metastasis and prognosis of primary cSCC. Funding: Sigrid Juselius Foundation, Finnish Cancer Research Foundation, Jane and Aatos Erkko Foundation, Cancer Foundation of the Southwest Finland, Finnish Dermatological Society, The Maud Kuistila Memorial Foundation, and Turku University Hospital. JSK is a doctoral candidate in the Doctoral Program for Clinical Investigation of the University of Turku. Declaration of Interest: None to declare. Ethical Approval: The study was approved by The Ethics Committee of the Hospital District of Southwest Finland (187/2006) and Auria Biobank’s Scientific Steering Committee (AB15-9721).