TCF-Net: A Hierarchical Transformer Convolution Fusion Network for Prostate Cancer Segmentation in Transrectal...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Accurate prostate segmentation in transrectal ultrasound (TRUS) is crucial for computer-aided diagnosis of prostate cancer.
• Challenges include image interferences, prostate shape variability, and limited datasets.

Purpose of the Study:

• To propose a novel Region-Adaptive Transformer Convolution Fusion Net (TCF-Net) for accurate and robust prostate segmentation in TRUS images.
• To address limitations of existing methods in handling interferences, shape variations, and small datasets.

Main Methods:

• Developed a hierarchical encoder-decoder network (TCF-Net) incorporating a region-adaptive transformer encoder and a convolution-based decoder.
• Implemented a patch-based fusion module to enhance fine prostate segmentation.
• Trained and evaluated the model on a clinical dataset of 1000 TRUS images from 135 patients.

Main Results:

• TCF-Net achieved a mean Intersection over Union (mIoU) of 94.4% on the challenging TRUS dataset.
• The proposed model outperformed other state-of-the-art methods by over 1% in segmentation accuracy.
• The region-adaptive transformer encoder effectively handled interferences and shape variations.

Conclusions:

• TCF-Net demonstrates superior performance in accurate and robust prostate segmentation from TRUS images.
• The model's design is effective in overcoming common challenges in medical image segmentation, particularly with limited data.
• TCF-Net shows significant potential for advancing computer-aided diagnosis of prostate cancer.