Integration of CT radiomics and clinical factors to predict pathological response to FLOT chemotherapy in gastric cancer

Aims: Neoadjuvant chemotherapy (NAC) has become a standard of care for locally advanced gastric cancer (LAGC), but a significant portion of patients do not respond to treatment, leading to unnecessary toxicity and delayed definitive surgery. Non-invasive, predictive tools are urgently needed to identify these non-responders prior to treatment. This study aimed to develop and validate a predictive model using an integrated approach of clinical factors and radiomic features extracted from pre-treatment computed tomography (CT) scans to classify LAGC patients as responders or non-responders to the neoadjuvant FLOT protocol. Methods: This retrospective study utilized 163 patients. A comprehensive feature set, including 11 patient biomarkers and 101 radiomic features, was used to train various machine learning algorithms via the pyCaret library. The Light Gradient Boosting Machine (LightGBM) was identified as the top-performing model. Results: On the internal validation, the LightGBM model achieved an accuracy of 80% with an AUC of 0.7926. Key features contributing to the model’s prediction included albumin, CA 19-9, age, and specific radiomic features such as Robust Mean Absolute Deviation. However, the model’s performance decreased in the independent external validation, yielding an accuracy of 54.59% and an AUC of 0.5409, highlighting challenges in generalizability. Conclusion: This study demonstrates the potential of an integrated clinical and radiomic approach using a LightGBM model to non-invasively predict treatment response to FLOT in LAGC. While the model showed strong internal performance, the decreased performance on an external validation set underscores the critical need for larger, multi-institutional, prospective studies with standardized data collection to establish clinical generalizability and widespread utility.