Towards U-Net-based intraoperative 2D dose prediction in high dose rate prostate brachytherapy

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Sub-optimal dosimetry in prostate high-dose-rate brachytherapy (HDR-BT) is often due to poor needle placement.
• Predicting the dosimetric impact of needle placement during HDR-BT is challenging, hindering widespread adoption of high-quality treatments.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting 2D dosimetry in prostate HDR-BT.
• To provide rapid, intra-operative feedback on needle implantation quality using axial transrectal ultrasound (TRUS) images.

Main Methods:

• Retrospective analysis of 248 prostate HDR-BT patient treatment plans.
• Training fifteen U-Net models to predict isodose levels (90%–200%) across prostate regions (base, midgland, apex).
• Evaluation using Dice Similarity Coefficient (DSC), precision, recall, surface distance, and Hausdorff distance, benchmarked against replanned cases.

Main Results:

• U-Net models achieved high accuracy, with median DSC of 0.97 for 90% isodose at midgland.
• Performance decreased with higher isodose levels and varied across prostate regions (e.g., median DSC 0.63–0.65 for 200% isodose).
• Median prediction time was rapid at 25 ms, suitable for real-time application.

Conclusions:

• U-Net models accurately predict HDR-BT isodose lines on 2D TRUS images with sufficient speed for intra-operative use.
• Integration of these auto-segmentation algorithms can provide real-time feedback on needle implantation quality.
• This technology has the potential to improve the quality and accessibility of prostate HDR-BT.