3D PBV-Net: An automated prostate MRI data segmentation method.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Healthcare

Background:

• Prostate cancer is a leading cause of male mortality worldwide.
• Accurate prostate segmentation in MRI is crucial for diagnosis and treatment planning.
• Existing automated methods struggle with image variability and low resolution.

Purpose of the Study:

• To develop an automated prostate MRI segmentation approach using bicubic interpolation with an improved 3D V-Net (3D PBV-Net).
• To enhance the accuracy and reliability of prostate gland segmentation in MRI data.

Main Methods:

• Preprocessing prostate MRI data using bicubic interpolation to address low-frequency components.
• Developing and applying a novel 3D PBV-Net model for segmentation.
• Evaluating the 3D PBV-Net on the PROMISE 12 and TPHOH clinical datasets.

Main Results:

• Achieved high average accuracy (97.65% and 98.29%) and Dice metrics (0.9613 and 0.9765).
• Demonstrated low Hausdorff distances (3.120 mm and 0.9382 mm) and average boundary distances (1.708 mm and 0.7950 mm).
• The 3D PBV-Net significantly improved automated segmentation accuracy.

Conclusions:

• The proposed 3D PBV-Net offers a promising solution for accurate automated prostate MRI segmentation.
• The method meets accuracy requirements for telehealth applications in prostate cancer care.