Deep learning performance on MRI prostate gland segmentation: evaluation of two commercially available algorithms...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Prostate Cancer Diagnostics

Background:

• Accurate whole-gland prostate segmentation is vital for advanced prostate cancer treatments like MRI-ultrasound fusion biopsy and radiotherapy.
• The number of commercially available AI deep learning algorithms (DLAs) for prostate segmentation is growing rapidly.
• Clinical validation of these AI models in real-world settings is often limited.

Purpose of the Study:

• To validate the performance of two commercially available, FDA-cleared and CE-marked DLAs for prostate gland segmentation.
• To assess AI model performance against expert radiologist manual segmentation using a heterogeneous clinical MRI dataset.
• To contribute to the understanding of AI model reliability in routine clinical practice.

Main Methods:

• Retrospective analysis of MRI data from 123 patients across 7 hospitals and 8 scanners (2 vendors, 1.5T/3T).
• Comparison of prostate segmentation contours generated by two DLAs (DLA1, DLA2) against an expert radiologist's manual contours (reference standard).
• Evaluation of segmentation accuracy using the Dice Similarity Coefficient (DSC) and paired t-tests, without prior in-house DLA training.

Main Results:

• The mean DSC for DLA1 against the expert reference standard was [insert DSC value].
• The mean DSC for DLA2 against the expert reference standard was [insert DSC value].
• Statistical analysis revealed no significant difference in DSC between DLA1 and DLA2 compared to the expert reference standard (p-value [insert p-value]).

Conclusions:

• Commercially available DLAs demonstrate accurate whole-gland prostate segmentation capabilities comparable to expert radiologists.
• These AI models show promise for reliable application in real-world clinical settings.
• Clinical implementation of validated AI segmentation tools can potentially optimize radiologist workflows and improve patient care.