Bladder Wall Segmentation in MRI Images via Deep Learning and Anatomical Constraints

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate bladder wall segmentation in MRI is crucial for early bladder tumor detection and diagnosis.
• Current segmentation methods face challenges due to weak boundaries and varied bladder shapes, often requiring manual parameter tuning and feature selection.

Purpose of the Study:

• To develop an automatic and accurate bladder wall segmentation method using deep learning and anatomical constraints.
• To overcome the limitations of manual adjustments and hand-crafted features in existing segmentation techniques.

Main Methods:

• An autoencoder was employed to model anatomical and semantic information of bladder walls, extracting low-dimensional feature representations from MRI and label images.
• These learned priors were integrated as constraints into a modified residual network to enhance segmentation plausibility.

Main Results:

• The proposed method demonstrated superior accuracy and reliability in segmenting bladder walls compared to related works.
• Experiments on 1092 MRI images yielded a Dice Similarity Coefficient (DSC) of 85.48%.

Conclusions:

• The deep learning-based method with anatomical constraints offers a more accurate and reliable solution for automatic bladder wall segmentation.
• This approach holds significant potential for improving the early detection and auxiliary diagnosis of bladder tumors through enhanced MRI analysis.