View in EDS

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

Saved in:
Bibliographic Details
Title: PSMA PET/MRI‐based Swin Transformer architecture for Gleason Score prediction in prostate cancer.
Authors: Yang, Tianshuo1 (AUTHOR), Zhang, Huai1 (AUTHOR), Peng, Huiling2 (AUTHOR), Niu, Xiaobing3 (AUTHOR), Yang, Fengjiao4 (AUTHOR), Zhang, Jun5 (AUTHOR), Wang, Qiuhu1 (AUTHOR), Fan, Junfu1 (AUTHOR), Song, Yaqi6 (AUTHOR) songaqi@163.com, Tao, Weijing1 (AUTHOR) weijingtao2021@vip.163.com
Source: Medical Physics. Jan2026, Vol. 53 Issue 1, p1-10. 10p.
Subjects: Prostate cancer, Gleason grading system, Magnetic resonance imaging, Data integration, Deep learning, Transformer models, Diagnostic imaging, Clinical decision making
Abstract: Background: Prostate cancer (PCa) management hinges on Gleason Score (GS) assessment, which currently requires invasive biopsies carrying risks of complications. This underscores the demand for noninvasive imaging alternatives. Purpose: This study aimed to develop a Swin Transformer‐based deep learning framework for noninvasive GS prediction in PCa, utilizing multi‐center PSMA PET/MRI data to support clinical decision‐making. Methods: This retrospective study included PCa patients with pathological GS who underwent PSMA PET and MRI scans from three centers. Patients were stratified into training, validation, and testing sets through stratified random sampling. PSMA PET and MRI scans were preprocessed through normalization, segmentation, and data augmentation. Our Swin Transformer architecture integrates a 3D patch embedding layer, four sequential Swin Transformer Blocks with shifted window attention mechanisms, and a multi‐layer perceptron (MLP) classification head. Performance was evaluated using AUC, accuracy, sensitivity, specificity, and precision. Results: A total of 225 PCa patients were included in our study. Compared to the PET and T2WI single‐modal approaches, the ADC‐based single‐modal model demonstrated superior performance across all metrics. The multimodal model based on PET, ADC and T2WI showed the best performance, with an AUC of 0.767, sensitivity of 0.722, specificity of 0.815, accuracy of 0.778, and precision of 0.722. Conclusion: The Swin Transformer model, leveraging multiparametric and multimodal PSMA PET/MRI data, provides an effective tool for noninvasive GS prediction and clinical decision‐making for PCa. Incorporating more data from additional institutions could enhance the model's generalizability and predictive accuracy. [ABSTRACT FROM AUTHOR]
Copyright of Medical Physics is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Background: Prostate cancer (PCa) management hinges on Gleason Score (GS) assessment, which currently requires invasive biopsies carrying risks of complications. This underscores the demand for noninvasive imaging alternatives. Purpose: This study aimed to develop a Swin Transformer‐based deep learning framework for noninvasive GS prediction in PCa, utilizing multi‐center PSMA PET/MRI data to support clinical decision‐making. Methods: This retrospective study included PCa patients with pathological GS who underwent PSMA PET and MRI scans from three centers. Patients were stratified into training, validation, and testing sets through stratified random sampling. PSMA PET and MRI scans were preprocessed through normalization, segmentation, and data augmentation. Our Swin Transformer architecture integrates a 3D patch embedding layer, four sequential Swin Transformer Blocks with shifted window attention mechanisms, and a multi‐layer perceptron (MLP) classification head. Performance was evaluated using AUC, accuracy, sensitivity, specificity, and precision. Results: A total of 225 PCa patients were included in our study. Compared to the PET and T2WI single‐modal approaches, the ADC‐based single‐modal model demonstrated superior performance across all metrics. The multimodal model based on PET, ADC and T2WI showed the best performance, with an AUC of 0.767, sensitivity of 0.722, specificity of 0.815, accuracy of 0.778, and precision of 0.722. Conclusion: The Swin Transformer model, leveraging multiparametric and multimodal PSMA PET/MRI data, provides an effective tool for noninvasive GS prediction and clinical decision‐making for PCa. Incorporating more data from additional institutions could enhance the model's generalizability and predictive accuracy. [ABSTRACT FROM AUTHOR]
ISSN:00942405
DOI:10.1002/mp.70274