Artificial Analysis Applied to Treatment of Granulosa Cell Tumors of the Ovary

OPhir Nave^1*Pnina Barasheshet²

• ¹Jerusalem College of Technology, Jerusalem, Israel
• ²Ben-Gurion University of the Negev, Be'er Sheva, Israel

We propose a novel method that integrates a mathematical model describing the combined action of TRAIL-producing oncolytic virus and PAC-1 therapy for granulosa cell tumors of the ovary with machine learning (ML) algorithms to enhance the prediction of tumor dynamics. This approach follows a structured approach using a few stages. First, the ML algorithm is trained to predict tumor burden based on clinico-pathological data and imaging findings collected at multiple time points during treatment. Second, the mathematical model is personalized for individual patients, incorporating initial tumor characteristics. Third, critical variables from the model solutions are extracted and added as new features for ML training. Finally, various ML algorithms are applied to the enriched dataset. Our comparative analysis showed that integrating the mathematical model reduced the root mean square error in linear regression predictions and improved both accuracy and F1-scores in neural network models. These results were confirmed across distinct patient cohorts treated with TRAIL-oncolytic virus and PAC-1 therapy.