Deep-learning-based segmentation using individual patient data on prostate cancer radiation therapy.

Area of Science:

• Medical Physics
• Radiotherapy
• Medical Imaging

Background:

• Organ-at-risk segmentation is crucial for adaptive radiotherapy (ART).
• Automated segmentation using deep learning can enhance efficiency and reduce manual labor in ART.
• Prostate cancer treatment requires precise delineation of organs at risk.

Purpose of the Study:

• To develop an auto-segmentation model for organs at risk in prostate cancer patients.
• To utilize individual patient datasets and deep learning-based augmentation for personalized ART.
• To tailor radiation therapy based on anatomical changes during treatment.

Main Methods:

• A deep learning-based augmentation method using deformable vector fields (DVFs) was employed.
• An nnU-net autosegmentation network was trained on augmented CT images.
• Patient-specific models were created and evaluated using Dice Similarity Coefficient (DSC), Hausdorff distance, and mean surface distance.

Main Results:

• Patient-specific auto-segmentation models were successfully developed.
• High DSC values were achieved for bladder (0.94 ± 0.03), prostate (0.84 ± 0.07), and rectum (0.83 ± 0.04).
• The model demonstrated accuracy comparable to models trained on large datasets.

Conclusions:

• The feasibility of automatic segmentation using individual patient data and augmentation techniques was demonstrated.
• The proposed method shows potential for clinical application in automatic prostate segmentation for ART.
• This approach can personalize radiation therapy for prostate cancer patients.