Deep dense multi-path neural network for prostate segmentation in magnetic resonance imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate prostate segmentation in Magnetic Resonance (MR) images is crucial for diagnosis and treatment planning.
• Existing segmentation methods may lack robustness and accuracy across diverse imaging conditions.

Purpose of the Study:

• To develop and evaluate a 3D deep dense multi-path convolutional neural network for automated prostate segmentation in MR images.

Main Methods:

• An encoder-decoder based 3D convolutional neural network architecture with densely connected layers for feature extraction.
• Utilized residual layout and grouped convolution in the decoder for precise volume prediction.
• Trained the network using sub-volumes of MR images centered on the prostate.

Main Results:

• Achieved high Dice coefficients (95.11% and 89.01%) on two independent datasets.
• Demonstrated robust segmentation performance despite variations in MR image quality.
• Qualitative and quantitative comparisons showed comparable or superior performance to existing approaches.

Conclusions:

• The proposed 3D convolutional neural network effectively segments the prostate with high accuracy and robustness.
• This method holds potential for application in segmenting other types of medical images.