Anatomy-aware Fine-grained Multimodal Fusion for Laryngopharyngeal Cancer T-Staging Prediction Using CT and Radiology...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate T-staging is vital for personalized treatment of laryngopharyngeal cancer.
• Current methods like invasive biopsies and CT-based staging have limitations in precision and complexity.
• Existing computer-aided methods struggle with modeling anatomical invasion patterns and aligning imaging with textual data.

Purpose of the Study:

• To develop an anatomy-aware multimodal framework for improved laryngopharyngeal cancer T-staging.
• To address challenges in structural relationship modeling and cross-modal alignment in current staging methods.

Main Methods:

• Proposed an anatomy-aware multimodal framework integrating CT scans and radiology reports.
• Constructed an Anatomy-Structured Organ Graph (AOG) to model invasion patterns.
• Implemented Organ-Anchored Cross-Modal Alignment (OCA) and Report-Enhanced Graph-Refinement (REG) for unified representation.

Main Results:

• The proposed framework achieved superior performance in laryngopharyngeal cancer T-staging.
• Demonstrated effective integration of spatial CT information and textual evidence from radiology reports.
• Successfully modeled complex tumor invasion patterns and aligned organ-specific details.

Conclusions:

• The novel multimodal framework significantly enhances T-staging accuracy for laryngopharyngeal cancer.
• This approach offers a promising non-invasive alternative to traditional staging methods.
• The integration of anatomical context and textual data represents a significant advancement in cancer staging AI.