Predicting Chemotherapy-Induced Peripheral Neuropathy Using Transformer-Based Multimodal Deep Learning.

Area of Science:

• Oncology
• Computational Biology
• Medical Informatics

Background:

• Chemotherapy-induced peripheral neuropathy (CIPN) significantly impacts cancer patient quality of life and treatment adherence.
• Existing prediction models for CIPN using single-modal data lack sufficient accuracy for clinical application.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting CIPN by integrating multimodal patient data.
• To enhance the accuracy and interpretability of CIPN prediction models for clinical utility.

Main Methods:

• A Transformer-based deep learning architecture was employed for intermediate data fusion.
• Multimodal data including clinical, genomic, biosignal, wearable, and imaging information were integrated from EHRs and public databases.
• Model interpretability was achieved using SHAP and Grad-CAM, with performance evaluated by AUC-ROC, accuracy, sensitivity, specificity, and F1-score.

Main Results:

• The Transformer model achieved superior performance (AUC=0.93, accuracy=88.5%) compared to conventional models.
• Key predictors identified include chemotherapy dosage, nerve MRI, ECG changes, CYP2C8 mutations, and diabetes.
• High predicted CIPN risk correlated with significantly lower overall survival, indicating broader systemic implications.

Conclusions:

• Deep learning models integrating multimodal data significantly improve CIPN prediction accuracy.
• Explainable AI techniques support the clinical implementation of these models in precision oncology.
• Future work should focus on multicenter validation and developing neuroprotective strategies for at-risk patients.