Predicting complete response to concurrent chemoradiotherapy in locally advanced cervical squamous cell carcinoma...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Cervical cancer remains a significant global health concern.
• Concurrent chemoradiation is standard for locally advanced squamous cell carcinoma, but 20-30% of patients experience persistent disease.
• Predicting treatment response is crucial for improving patient outcomes.

Purpose of the Study:

• To develop a predictive model for persistent cervical cancer.
• Utilize pretreatment multisequence magnetic resonance imaging (MRI) data.
• Employ advanced deep learning techniques, including Crossformer, for prediction.

Main Methods:

• Retrospective study of 259 patients with locally advanced cervical squamous cell carcinoma.
• Generation of 2.5D data from four MRI sequences.
• Development and comparison of a deep learning model (Crossformer) against radiomics and clinical models.

Main Results:

• The CrossFormer deep learning model achieved high predictive accuracy (AUC 0.884 training, 0.833 validation, 0.814 test).
• It outperformed traditional convolutional neural network models and radiomics/clinical models.
• Combining deep learning with clinical features further enhanced predictive performance (AUC up to 0.914).

Conclusions:

• A deep learning model using 2.5D multi-sequence MRI shows strong predictive performance for persistent cervical cancer.
• This approach offers a promising, clinically applicable tool for treatment decision-making in locally advanced cervical squamous cell carcinoma.
• Early identification of persistent disease can guide subsequent management strategies.