A Graph Attention Network-Based Multimodal Auxiliary Intelligent Grading Model for Uterine Prolapse Severity

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Pelvic Floor Disorders

Background:

• Uterine prolapse significantly impacts women's quality of life.
• Current diagnostic methods for uterine prolapse have limited accuracy.
• Existing multimodal data fusion techniques lack effective cross-modal strategies.

Purpose of the Study:

• To develop and validate the Multimodal Aggregation Graph Attention Network (MAGANet) for objective and efficient uterine prolapse grading.
• To enhance diagnostic accuracy by fusing clinical features and pelvic floor MRI data.
• To address limitations in current cross-modal fusion strategies for uterine prolapse diagnosis.

Main Methods:

• Retrospective collection of clinical and pelvic floor MRI data from 564 patients.
• Development of MAGANet with modules for data extraction, feature fusion, and graph representation.
• Utilized a Multi-level Gated Graph Attention Network for grading results based on fused multimodal data.

Main Results:

• MAGANet achieved high accuracy on an independent test set: ACC 0.935, AUC 0.980.
• The model demonstrated superior performance compared to single-modal approaches and other fusion methods.
• Key performance metrics included precision 0.868, macro-F1 0.908, kappa 0.803, sensitivity 0.962, and specificity 0.650.

Conclusions:

• MAGANet effectively integrates multimodal data for superior uterine prolapse grading.
• The model offers an objective and efficient tool for clinical diagnosis.
• This study provides new insights into multimodal intelligent diagnosis for pelvic floor disorders.