Automatic AI-based contouring of prostate MRI for online adaptive radiotherapy

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• MR-guided radiotherapy (MRgRT) enables daily plan adaptation, crucial for pelvic tumors due to organ variability.
• Current MRgRT workflows lack fast automatic annotation, hindering efficient online adaptation.
• Adaptive MRgRT can significantly benefit patients with pelvic tumors by sparing organs at risk.

Purpose of the Study:

• To train and validate a deep learning model for rapid, accurate automatic MRI segmentation.
• To enable the implementation of automatic contouring in clinical online MRgRT workflows.
• To improve the efficiency and precision of adaptive radiotherapy for pelvic cancers.

Main Methods:

• A deep learning model was trained on 232 T2w MRI datasets from 47 patients undergoing MR-Linac treatment.
• Manual annotations included prostate, seminal vesicles, rectum, bladder, and bony structures.
• The model was validated on 20 unseen MRIs, with quantitative (DSC, HD) and qualitative (physician scoring) evaluations.

Main Results:

• The AI model demonstrated high accuracy, with Dice Similarity Coefficients (DSC) up to 0.97 for the bladder.
• Quantitative metrics showed acceptable performance for organs like the rectum (median 95% HD of 6.9 mm).
• Qualitative evaluation revealed 80% of contours were clinically acceptable, and 16% required only minor adjustments.

Conclusions:

• An AI-based autocontouring model was successfully developed and validated for online adaptive MR-guided radiotherapy.
• The model generates contours that are largely accepted by physicians or require minimal corrections.
• This AI tool is suitable for clinical implementation in MRgRT workflows for prostate cancer treatment.