Fully automated localization of prostate peripheral zone tumors on apparent diffusion coefficient map MR images using...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Accurate detection and localization of prostate cancer (PCa) are crucial for effective targeted biopsies and treatment planning.
• Fully automated localization of peripheral zone (PZ) PCa using apparent diffusion coefficient (ADC) maps could significantly enhance clinical utility.

Purpose of the Study:

• To develop and describe an automated method for localizing PCa within the PZ on ADC map MR images.
• To utilize an ensemble U-Net-based model for precise PCa detection and segmentation.

Main Methods:

• A retrospective case-control study involving 226 patients (154 with and 72 without clinically significant PZ PCa).
• An ensemble U-Net model was trained and tested on 3T ADC map MR images.
• Ground truth was established by expert radiologists comparing manual delineations with MRI-radical prostatectomy maps.

Main Results:

• The developed algorithm achieved a Dice Similarity Coefficient (DSC) of 86.72% ± 9.93%, sensitivity of 85.76% ± 23.33%, and specificity of 76.44% ± 23.70% on 80 test cases.
• The area under the receiver operating characteristic curve (AUC) was reported as 0.779.
• The ensemble model demonstrated high performance in detecting and segmenting PCa in the PZ.

Conclusions:

• An ensemble U-Net-based approach provides accurate detection and segmentation of PCa in the PZ from ADC map MR prostate images.
• This automated method holds potential for clinical application in prostate cancer management.