PSMA PET/MRI-based Swin Transformer architecture for Gleason Score prediction in prostate cancer

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Cancer Diagnosis
• Radiomics and Quantitative Imaging

Background:

• Prostate cancer (PCa) management relies on Gleason Score (GS) assessment.
• Current GS assessment requires invasive biopsies with associated risks.
• There is a significant need for noninvasive diagnostic methods in PCa.

Purpose of the Study:

• To develop a Swin Transformer-based deep learning framework.
• To predict Gleason Score (GS) noninvasively in prostate cancer (PCa).
• To utilize multi-center PSMA PET/MRI data for enhanced clinical decision-making.

Main Methods:

• Retrospective study of 225 PCa patients with pathological GS.
• Utilized multi-center PSMA PET and MRI scans.
• Developed a Swin Transformer architecture with 3D patch embedding and shifted window attention for multimodal data integration.

Main Results:

• The multimodal model integrating PET, ADC, and T2WI achieved the best performance.
• Achieved an AUC of 0.767, sensitivity of 0.722, specificity of 0.815, accuracy of 0.778, and precision of 0.722.
• Outperformed single-modal approaches, particularly ADC-based models.

Conclusions:

• The Swin Transformer model effectively predicts Gleason Score noninvasively using multimodal PSMA PET/MRI data.
• This AI tool supports clinical decision-making for prostate cancer (PCa).
• Further validation with larger, multi-institutional datasets can improve generalizability and accuracy.