Related Experiment Video
Updated: May 1, 2026
15:48
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
22.3K
Z-SSMNet: Zonal-aware Self-supervised Mesh Network for prostate cancer detection and diagnosis with Bi-parametric MRI
Yuan Yuan1, Euijoon Ahn2, Dagan Feng3
1School of Computer Science, Faculty of Engineering, The University of Sydney, Sydney, 2006, NSW, Australia.
Summary
A new AI system, Zonal-aware Self-supervised Mesh Network (Z-SSMNet), improves prostate cancer detection using bi-parametric MRI. It addresses limitations in current AI by better learning spatial information and using self-supervised learning on large datasets.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Bi-parametric MRI (bpMRI) is crucial for detecting clinically significant prostate cancer (csPCa).
- Current AI methods struggle with anisotropic bpMRI data and require extensive annotated datasets.
- Limitations exist in learning spatial information from bpMRI using standard CNNs and Transformers.
Purpose of the Study:
- To develop an AI system for improved csPCa detection and diagnosis using bpMRI.
- To overcome limitations of existing AI models in handling anisotropic bpMRI data and dataset scarcity.
- To enhance the efficiency and cost-effectiveness of prostate cancer management through AI.
Main Methods:
- Proposed the Zonal-aware Self-supervised Mesh Network (Z-SSMNet) integrating multi-dimensional convolutions.
- Implemented a self-supervised learning (SSL) technique for capturing intra-slice and inter-slice semantic information from unlabeled data.
- Constrained the network to focus on zonal anatomical regions for improved csPCa detection.
Main Results:
- Z-SSMNet achieved top performance in the PI-CAI challenge's Open Development Phase (AP: 0.633, AUROC: 0.881).
- The model secured second place in the Closed Testing Phase (AP: 0.690, AUROC: 0.909).
- Demonstrated superior lesion-level detection and patient-level diagnosis capabilities on a large, multi-center dataset.
Conclusions:
- Z-SSMNet shows significant potential for AI-driven csPCa diagnosis and management.
- The proposed methods effectively address challenges in analyzing anisotropic bpMRI data.
- AI systems utilizing advanced techniques can enhance prostate cancer care.
