Automatic prostate segmentation on MR images with deep network and graph model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Prostate cancer has a high mortality rate, necessitating accurate automated diagnosis and treatment.
• Current prostate segmentation methods struggle with blurred boundaries and lack of intrinsic structural feature extraction.
• Unsupervised prostate segmentation remains a challenging research area in medical image analysis.

Purpose of the Study:

• To develop a novel automated prostate segmentation model utilizing deep learning for improved accuracy.
• To overcome limitations of conventional methods relying on handcrafted features for prostate segmentation.
• To enhance the extraction of intrinsic prostate structures, particularly in images with ambiguous boundaries.

Main Methods:

• A deep network is employed to learn features for precise prostate boundary refinement.
• Prostate proposals are generated in the transverse plane using position and shape estimation.
• A graph-based approach considers correlations across sequential images for 3D segmentation refinement.

Main Results:

• The proposed deep network and graph-based method significantly outperforms existing state-of-the-art techniques.
• Superior performance is demonstrated using Dice Similarity Coefficient and Hausdorff Distance metrics.
• The model shows effectiveness on a public dataset for automated prostate segmentation.

Conclusions:

• The developed deep learning and graph-based model offers a superior solution for automated prostate segmentation.
• This approach effectively addresses challenges posed by blurred boundaries and complex image structures.
• The findings contribute to advancing automated prostate cancer diagnosis and treatment planning.