Prediction of Neoadjuvant Chemotherapy Efficacy for Locally Advanced Nasopharyngeal Carcinoma Using MRI-Based Deep...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Accurate prediction of neoadjuvant chemotherapy (NACT) efficacy is critical for managing advanced nasopharyngeal carcinoma (LA-NPC).
• Current predictive models often lack generalizability, necessitating novel approaches.
• The combined use of MRI-based deep learning features (DLF) and Vision Transformer (ViT) for this application is unexplored.

Purpose of the Study:

• To evaluate the predictive value of multi-sequence MRI-based DLF integrated with ViT for NACT efficacy in LA-NPC.
• To compare the performance of this integrated model against traditional radiomics and standalone deep learning models.

Main Methods:

• Retrospective analysis of 266 LA-NPC patients treated with NACT.
• Development and comparison of traditional radiomics, 2D, 2.5D, and 3D deep learning models.
• Extraction of DLF from optimal models, dimensionality reduction (PCA), and input into a ViT architecture.

Main Results:

• The best traditional radiomics model (XGBoost on T2-FS) achieved an AUC of 0.760.
• 3D deep learning models demonstrated superior performance (best AUC: 0.755) compared to 2D and 2.5D models.
• The integrated DLF-ViT model achieved a significantly higher validation AUC of 0.926, accuracy of 0.903, and F1-score of 0.927.

Conclusions:

• The integrated model combining multi-sequence MRI DLF with ViT significantly enhances predictive performance for NACT efficacy in LA-NPC.
• This novel approach offers a more accurate and generalizable tool for treatment response prediction in LA-NPC.