A Multi-Modal Deep Learning Approach for Predicting Eligibility for Adaptive Radiation Therapy in Nasopharyngeal...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Adaptive radiation therapy (ART) improves nasopharyngeal carcinoma (NPC) prognosis.
• Patient selection for ART is challenging due to anatomical variability, impacting treatment duration and radiologist workload.

Purpose of the Study:

• To predict eligible ART candidates for NPC patients before radiation therapy (RT) using a classification neural network.
• To leverage fused medical imaging and clinical data for efficient ART candidate selection.

Main Methods:

• Retrospective data from 305 NPC patients undergoing RT were analyzed.
• A multi-modal classification neural network combining ResNet-50, cross-attention, multi-scale features, and clinical data was developed.
• Patients were classified based on ART re-planning status.

Main Results:

• The proposed multi-modal deep prediction model achieved an AUC of 0.9070.
• The model demonstrated superior performance compared to other deep learning networks in predicting ART eligibility.
• Accurate classification and prediction of ART eligibility for NPC patients were achieved.

Conclusions:

• The developed method shows strong performance in predicting ART eligibility for NPC patients.
• This approach has the potential to enhance clinical decision-making and optimize treatment efficiency.
• The findings support the advancement of personalized cancer care through AI-driven predictions.