Automatic prostate segmentation of magnetic resonance imaging using Res-Net

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Prostate segmentation from MRI is crucial for cancer diagnosis and treatment response evaluation.
• Challenges include unclear boundaries, tissue heterogeneity, shape variations, and limited annotated data.

Purpose of the Study:

• To develop a novel two-stage segmentation method for automatic prostate segmentation.
• To achieve accurate and reproducible results using multi-site and multi-vendor datasets.

Main Methods:

• A two-stage neural network approach combining U-Net and residual blocks.
• The first stage uses 2D U-Net for approximate localization.
• The second stage employs U-Net with Res-Net for precise segmentation.

Main Results:

• Achieved an average Dice Similarity Coefficient (DSC) of 93.8%.
• Obtained Sensitivity of 94.6% and Specificity of 99.3% on test datasets.
• The network was trained on 116 patient datasets with an 80/10/10 split for training/validation/testing.

Conclusions:

• The proposed two-stage neural network significantly enhances prostate segmentation accuracy.
• This method supports more accurate and reproducible prostate cancer diagnosis and treatment evaluation.