Cascade Path Augmentation Unet for bladder cancer segmentation in MRI.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Oncology research

Background:

• Accurate segmentation of bladder cancer (BCa) structures, including inner wall (IW), outer wall (OW), and bladder tumor (BT), is vital for diagnosing muscle-invasive bladder cancer (MIBC).
• Current methods require precise delineation for effective computer-aided diagnosis.

Purpose of the Study:

• To introduce a novel deep learning model, CPA-Unet, designed to enhance the segmentation accuracy of IW, OW, and BT.
• To provide a tool beneficial for clinical applications in bladder cancer diagnosis.

Main Methods:

• A Cascade Path Augmentation Unet (CPA-Unet) was developed for multi-regional bladder segmentation using 1545 T2-weighted MRI scans.
• The model utilizes a cascade strategy for background elimination, U-Net for initial segmentation, path augmentation for multi-scale feature extraction, and partial dense connections for feature fusion.

Main Results:

• The CPA-Unet demonstrated superior performance in segmenting IW, OW, and BT compared to existing deep learning methods.
• Achieved high Dice Similarity Coefficients (DSC) and low Hausdorff Distances (HD): IW (DSC: 98.19%, HD: 2.07 mm), OW (DSC: 82.24%, HD: 2.62 mm), and BT (DSC: 87.40%, HD: 0.76 mm).

Conclusions:

• The CPA-Unet effectively segments the bladder's inner and outer walls, along with tumors.
• The model's accurate segmentation provides a reliable basis for computer-assisted clinical diagnosis of MIBC.