Recurrence risk stratification for locally advanced cervical cancer using multi-modality transformer network

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Oncology
• Cervical Cancer Research

Background:

• Recurrence risk stratification is crucial for locally advanced cervical cancer (LACC) management.
• Accurate prediction of recurrence aids in personalized treatment strategies and improved patient outcomes.

Purpose of the Study:

• To evaluate the efficacy of a transformer network for recurrence risk stratification in LACC patients.
• To compare the performance of the transformer network against conventional radiomics and other deep learning models.

Main Methods:

• A cohort of 104 LACC patients underwent CT and MR imaging.
• A multi-modality transformer network was developed to extract multi-modal and multi-scale features for recurrence prediction.
• Model performance was evaluated using metrics including AUC, accuracy, f1-score, sensitivity, specificity, and precision.

Main Results:

• The transformer network achieved a high AUC of 0.819 ± 0.038 in the testing cohort.
• The proposed network outperformed conventional radiomics methods (AUCs ranging from 0.680 to 0.777) and other deep learning networks (AUCs of 0.743 and 0.733).
• Consistent superior performance was observed across training, validation, and testing cohorts.

Conclusions:

• The multi-modality transformer network demonstrates significant promise for recurrence risk stratification in LACC.
• This AI-driven approach can serve as a valuable tool to support clinical decision-making for LACC patients.
• Further integration of such advanced imaging analysis techniques can enhance precision oncology for cervical cancer.