STASis-Scan

STASis-Scan was created in response to a frustrating clinical reality in lung cancer care. Surgeons often perform surgery without knowing whether STAS is present, even though it is one of the strongest predictors of recurrence, and it is usually discovered only after the operation when the surgical approach can no longer be changed. Our goal was to bridge the gap between research and clinical decision making by translating findings from more than twenty seven peer reviewed studies into an interactive platform that helps explore recurrence risk and surgical strategy. To build this system we combined insights from molecular biology, including epithelial mesenchymal transition markers and TP53 enrichment patterns, with machine learning to train a Random Forest model that achieves about eighty three percent accuracy, with STAS emerging as the most influential predictor. One of the biggest challenges was the lack of a unified public dataset containing clinical variables, molecular biomarkers, and surgical outcomes together, which required constructing literature calibrated synthetic data while ensuring the relationships matched published studies. We also encountered deployment and technical challenges while preparing the application for web hosting and validating that the model behaved consistently with clinical expectations. We are proud that the final prototype translates a large body of research into an accessible interface that allows users to compare recurrence risk between lobectomy and segmentectomy in real time. Through this project we learned how complex it is to bridge biomedical research with usable decision support tools, and how integrating tumor biology with computational modeling can provide new ways to explore surgical risk. In the future, STASis-Scan could be expanded through validation on real clinical datasets and collaboration with thoracic oncology researchers to further refine and evaluate the model.

Built With

• python