Machine Learning for Preoperative Assessment and Postoperative Prediction in Cervical Cancer: Multicenter...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Machine learning (ML) is increasingly used in cervical cancer (CC) research, but few studies integrate clinical and imaging data.
• Accurate preoperative assessment of parametrial invasion and lymph node metastasis, and postoperative prognosis prediction are crucial for CC patient management.

Purpose of the Study:

• To develop an integrated ML model combining clinicopathological variables and MRI features.
• To enhance preoperative detection of parametrial invasion and lymph node metastasis in CC.
• To improve postoperative prediction of recurrence and survival for CC patients.

Main Methods:

• Retrospective analysis of 250 CC patients' data (2014-2022).
• Evaluation of 7 ML models (KNN, SVM, DT, RF, balanced RF, weighted DT, weighted KNN) for predictive performance.
• Performance assessment using 5-fold cross-validation (accuracy, sensitivity, specificity, precision, F1-score, AUC).

Main Results:

• Balanced Random Forest (RF) showed optimal performance for preoperative parametrial invasion detection (sensitivity 0.81, specificity 0.85).
• Weighted K-nearest neighbor (KNN) excelled in lymph node metastasis detection (sensitivity 0.98, AUC 0.72) and postoperative prognosis prediction (recurrence accuracy 0.94, AUC 0.86; mortality accuracy 0.97, AUC 0.77).
• An AI-assisted system was developed for preoperative and postoperative predictions.

Conclusions:

• Integrating clinical data and MRI features enhances diagnostic and prognostic capabilities for CC.
• This approach facilitates personalized and precise treatment strategies for cervical cancer patients.
• The developed AI system supports improved preoperative evaluation and postoperative prognosis prediction.