Related Experiment Video
Updated: Aug 5, 2026
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
CSFBNet: Cosine-Consistency Filtering and Dual-Stream Complementary Semantic Guidance for Prostate Cancer MRI
IEEE Journal of Biomedical and Health Informatics
|July 31, 2026
Summary
This study introduces CSFBNet, a novel prostate MRI segmentation network. CSFBNet improves accuracy by filtering features and enhancing lesion cues, offering a practical solution for prostate cancer detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate prostate and prostate cancer segmentation on MRI is difficult due to small structures, low contrast, blurred boundaries, and surrounding tissue interference.
- Existing methods struggle with precise delineation, impacting diagnostic reliability and treatment planning.
Purpose of the Study:
- To develop and evaluate CSFBNet, a deep learning segmentation network designed to overcome the challenges in prostate MRI segmentation.
- To improve the accuracy and robustness of automated prostate and prostate cancer segmentation using advanced feature processing techniques.
Main Methods:
- CSFBNet employs a novel architecture combining cosine-consistency-based feature filtering (CSSB) and foreground-background complementary semantic guidance (FBAC).
- The CSSB block refines shallow features and minimizes background noise, while the FBAC block enhances lesion-specific information during decoding.
- The model was validated on three diverse datasets: PROMISE12, HY Prostate, and PI-CAI.
Main Results:
- CSFBNet achieved high Dice scores on the PROMISE12 (0.9017) and HY Prostate (0.6539) datasets, showing significant improvements over the baseline model.
- The network demonstrated competitive performance in region-overlap and false-positive suppression compared to other state-of-the-art segmentation methods.
- While effective, challenges remain in small-lesion sensitivity and boundary recovery, particularly noted in the PI-CAI dataset.
Conclusions:
- CSFBNet presents a practical and effective framework for automated prostate MRI segmentation, enhancing diagnostic capabilities.
- The proposed method improves region-overlap accuracy and reduces false positives, contributing to more reliable prostate cancer assessment.
- Future research should focus on enhancing sensitivity to small lesions and improving boundary delineation in complex cases.