Related Experiment Video
Updated: Jun 5, 2025
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
135
Cross-Slice Attention and Evidential Critical Loss for Uncertainty-Aware Prostate Cancer Detection
Alex Ling Yu Hung1,2, Haoxin Zheng1,2, Kai Zhao1
1Department of Radiological Science, UCLA.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|December 16, 2024
Summary
This study introduces a new 2.5D deep learning model for detecting prostate cancer in MR images. The model achieves state-of-the-art results and provides better uncertainty estimates for clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Current deep learning models for medical image analysis often ignore volumetric data or struggle with anisotropic MR data resolution.
- Accurate uncertainty estimation in AI predictions is crucial for clinical decision-making.
Purpose of the Study:
- To develop a novel 2.5D cross-slice attention model for improved prostate cancer detection in MR images.
- To enhance the model's ability to provide reliable epistemic uncertainty estimation.
Main Methods:
- A novel 2.5D cross-slice attention model integrating global and local information was developed.
- Evidential deep learning was employed, utilizing an evidential critical loss function.
- The model was evaluated on two distinct datasets for prostate cancer detection.
Main Results:
- The proposed model achieved state-of-the-art performance in prostate cancer detection.
- The model demonstrated improved epistemic uncertainty estimation compared to existing methods.
- Extensive experiments validated the model's effectiveness on multiple datasets.
Conclusions:
- The 2.5D cross-slice attention model offers a significant advancement for prostate cancer detection in MR imaging.
- The integration of evidential deep learning and critical loss enhances prediction reliability and uncertainty quantification.
- This approach holds promise for improving diagnostic accuracy and supporting clinical decisions in men's health.
