Related Experiment Video
Updated: May 10, 2025
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
106
Assessing Cancer Presence in Prostate MRI Using Multi-Encoder Cross-Attention Networks
Avtantil Dimitriadis1,2,3, Grigorios Kalliatakis1, Richard Osuala2
1Institute of Computer Science, Foundation for Research and Technology Hellas (FORTH), N. Plastira 100, Vassilika Vouton, 70013 Heraklion, Greece.
Journal of Imaging
|April 25, 2025
Summary
This study introduces a novel deep learning model for prostate cancer (PCa) diagnosis using bi-parametric MRI. The model accurately distinguishes PCa presence from non-cancerous findings, achieving a 0.91 AUC.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Prostate cancer (PCa) is a leading cause of cancer death in men, necessitating accurate diagnostic methods.
- Current diagnostic approaches for PCa using magnetic resonance imaging (MRI) often focus on lesion detection or classification, not the presence of cancer itself.
- Distinguishing between PCa and benign conditions from MRI is a critical but underexplored step in the clinical workflow.
Purpose of the Study:
- To develop and evaluate a deep learning model for the first time on a large-scale dataset to differentiate between the presence and absence of prostate cancer using MRI.
- To address the gap in current research by focusing on the critical task of cancer presence determination in the PCa clinical workflow.
Main Methods:
- Utilized the multi-centric ProstateNET Imaging Archive, the largest collection of PCa MR images, comprising over 6 million representations from 11,000+ cases.
- Trained a multi-encoder-cross-attention-fusion architecture on bi-parametric MR (bpMRI) images and clinical variables from 4504 patients.
- Evaluated the model's performance on two independent test sets: 975 retrospective and 435 prospective patients.
Main Results:
- Achieved a high area under the receiver operating characteristic curve (AUC) of 0.91, indicating strong diagnostic performance.
- Demonstrated the model's capability to effectively fuse information from complex bi-parametric imaging modalities.
- Showcased enhanced model robustness in distinguishing PCa presence across diverse patient cohorts.
Conclusions:
- The developed deep learning model shows significant promise for accurately determining the presence of prostate cancer from MRI.
- The method's ability to fuse multi-modal imaging data enhances robustness, supporting its potential for clinical adoption.
- This work paves the way for improved diagnostic accuracy and reduced mortality through advanced AI in oncology.
