Deep learning-based three-dimensional segmentation of the prostate on computed tomography images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Prostate segmentation on CT scans is crucial for treatment planning but is hindered by low soft-tissue contrast, leading to time-consuming manual delineation and high interobserver variability.
• Accurate segmentation is essential for effective radiation therapy and other prostate interventions.

Purpose of the Study:

• To develop and evaluate an automated, 3D prostate segmentation algorithm for CT images.
• To compare the performance of the automated algorithm against manual segmentation by an expert radiologist.

Main Methods:

• A customized U-Net convolutional neural network architecture was employed for 3D segmentation.
• The model was trained and validated on 69 3D abdominal CT scans and tested on a separate set of 23 scans.

Main Results:

• The automated algorithm achieved high accuracy, demonstrated by a Dice Similarity Coefficient (DSC) of [insert DSC value], Mean Absolute Distance (MAD) of [insert MAD value] mm, and Signed Volume Difference (SVD) of [insert SVD value] %.
• The algorithm's performance surpassed the interobserver variability, which was [insert interobserver DSC value]% for DSC, [insert interobserver MAD value] mm for MAD, and [insert interobserver SVD value]% for SVD.

Conclusions:

• The developed algorithm provides a fast, accurate, and robust solution for 3D prostate segmentation on CT images.
• This automated approach has the potential to streamline treatment planning and improve consistency in prostate cancer management.