Development and Validation of a Deep Learning Radiomics Model Predicting Lymph Node Status in Operable Cervical...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate preoperative assessment of lymph node metastasis is crucial for cervical cancer staging and treatment planning.
• Current methods may have limitations in precisely identifying lymph node involvement.
• Deep learning and radiomics offer potential for improved non-invasive diagnostic capabilities.

Purpose of the Study:

• To develop and validate a deep learning radiomics model for predicting preoperative lymph node metastases in cervical cancer.
• To compare the performance of the deep neural network (DNN) model against traditional logistic regression and support vector machine models.
• To assess the model's accuracy and generalizability in independent test sets.

Main Methods:

• A cohort of 226 operable cervical cancer patients was retrospectively analyzed.
• Radiomic features were extracted from preoperative CT images, and five were selected based on predictive power.
• A deep neural network (DNN) model was developed, combining radiomic features with clinicopathological data (histology and grade).
• The DNN model's performance was validated internally and externally.

Main Results:

• The DNN model achieved an area under the curve (AUC) of 0.99 and 97.16% accuracy in internal validation.
• External validation demonstrated sustained performance with an AUC of 0.90 and 92.00% accuracy.
• The DNN model outperformed baseline logistic regression models in predicting lymph node status.

Conclusions:

• Deep learning radiomics provides a powerful, non-invasive tool for predicting lymph node metastases in cervical cancer.
• The developed model shows high accuracy and generalizability, aiding in preoperative risk stratification.
• This predictive model may assist in personalized treatment strategies, potentially favoring radiation therapy for select patients.