A new study demonstrates an artificial intelligence (AI) system designed to improve the treatment of cervical cancer by automatically outlining tumors on MRI scans and predicting the risk of local recurrence within five years. Accurate tumor outlining is essential for planning radiotherapy. Researchers developed a 3D deep-learning model called MedNext-7L to perform this task. The model showed slightly better accuracy and sensitivity than the commonly used nnUNetV2 system, meaning it can detect tumors more reliably and reduce the time doctors spend correcting treatment plans.
The study also used radiomics, a method that extracts detailed image features such as texture and shape that are not visible to the human eye. These features were analyzed using machine learning to predict whether the cancer would return within five years. When AI-generated tumor outlines were reviewed and quickly corrected by experts, the prediction accuracy was high (AUC 0.811). A fully automated approach without human correction had lower accuracy (AUC 0.730), but still showed promising results.
The findings suggest that this AI system could reduce doctors’ workload while helping identify high-risk patients who may not respond well to standard chemoradiotherapy. The study included 306 patients with locally advanced cervical cancer treated between 2011 and 2020, with follow-up lasting up to November 2024. Overall, the research highlights the potential for more efficient and personalized cancer treatment using AI.