Multimodal MRI Image Fusion for Early Automatic Staging of Endometrial Cancer

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Endometrial cancer (EC) staging is crucial for treatment planning.
• Accurate staging of early-stage EC (IA and IB) remains a clinical challenge.
• Current methods often involve complex segmentation followed by staging.

Purpose of the Study:

• To develop and validate a deep learning model for automated staging of early endometrial cancer using multimodal MRI fusion.
• To compare the diagnostic performance of the deep learning model against that of experienced radiologists.
• To assess the efficiency and precision of the proposed AI method compared to traditional staging techniques.

Main Methods:

• Retrospective analysis of 122 early EC patients with pathological confirmation.
• Utilized a Swin transformer model with a shift window multi-self-attention (SW-MSA) module for image analysis.
• Performed multimodal fusion of MRI images from sagittal, coronal, and transverse planes for classification.

Main Results:

• Individual MRI planes achieved high classification accuracy (sagittal: 0.988, coronal: 0.96, transverse: 0.94).
• Multimodal fusion of all three planes resulted in a perfect classification accuracy of 1.0.
• The automated Swin transformer method demonstrated 100% accuracy, recall, and specificity for early EC classification.

Conclusions:

• Multimodal MRI fusion with deep learning offers a highly accurate and automated approach for early endometrial cancer staging.
• The AI-driven method achieves diagnostic performance comparable to radiologists.
• This novel technique simplifies and enhances the precision of EC staging compared to prior segmentation-based methods.