Exhibit Hall | Forum 6
Purpose: AI modeling physicians’ clinical-decision-making (CDM) can improve the efficiency and accuracy of clinical practice or serve as a surrogate to provide initial consultations to patients seeking secondary opinions. In this study, we developed an AI network to model radiotherapy CDM and used dose prescription as an example to demonstrate its feasibility.
Methods: 96 patients with brain metastases treated by radiosurgery from 2017 to 2021 were included. CT images and tumor and organ-at-risk (OAR) contours were exported. Eight clinical parameters were extracted and digitized, including age, numbers of lesions, performance status (ECOG), symptomatic, surgery arrangement, re-treatment, primary cancer type, and metastasis to other sites. A 3D-CNN network was built using three encoding paths to capture three inputs: (1). tumor size, shape, and location; (2). the spatial relationship between tumor and OARs; (3). the clinical parameters. The model combines information from three paths to predict dose prescription. The actual prescription in the record was used as ground truth for model training. The model performance was assessed by 12-fold-cross-validation, with each fold consisting of randomly selected 80 training, 8 validation, and 8 testing subjects.
Results: The 96 patients included 48 low-dose prescriptions (1x20-22Gy) and 48 high-dose prescriptions (1x24Gy) from 8 physicians. 39(81%) low-dose and 44(92%) high-dose cases were prescribed correctly by the model. Among the 13 failed cases, 8 were caused by the physician practice variations, which were not accounted for in the group-data trained model. Including clinical parameters improved the overall prediction accuracy by 20%.
Conclusion: To our knowledge, this is the first study to demonstrate the feasibility of AI to predict dose prescription in CDM. Such CDM models can serve as vital tools to address healthcare disparities by providing preliminary consultations to patients in underdeveloped areas or as a valuable QA tool for physicians to cross-check intra- and inter-institution practices.
Funding Support, Disclosures, and Conflict of Interest: R01-EB028324, R01-EB001838, R01-CA226899
Not Applicable / None Entered.