Click here to

Session: Deep Learning for Image Quality in Treatment Planning [Return to Session]

Dosimetric and Image Quality Assessment of a Novel Deep Learning-Based Framework for MRI Synthesis From CT for Prostate High Dose-Rate Brachytherapy

A Podgorsak1,2*, B Venkatesulu1,2, M Abuhamad3, M Harkenrider1,2, A Solanki1,2, J Roeske1,2, H Kang1,2, (1) Department of Radiation Oncology, Stritch School of Medicine, Cardinal Bernadin Cancer Center, Loyola University Chicago, Maywood, IL, (2) Loyola University Medical Center, Maywood, IL, (3) Department of Computer Science, Loyola University Chicago, Chicago, IL


SU-H330-IePD-F5-1 (Sunday, 7/10/2022) 3:30 PM - 4:00 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 5

Purpose: The prostate delineated on MRI is reported to be more accurate compared with on CT. However, MRI acquisition slows the treatment workflow and the image registration process adds uncertainty to prostate high dose-rate (HDR) brachytherapy planning. To address these issues, MRI synthesis from CT for prostate HDR brachytherapy is investigated using a novel deep learning framework.

Methods: 58 paired CT and T2-weighted MRI datasets from patients treated with HDR brachytherapy at our institution comprised our training dataset. A novel generative adversarial network (GAN) was developed and trained for MRI synthesis from CT. The synthetic MRI (sMRI) data were assessed using 10 additional patient datasets from our institution. Line profiles were compared between the CT, real MRI (rMRI), and sMRI. sMRI peak signal-to-noise ratio (pSNR) and structural similarity index measure (SSIM) between the rMRI and sMRI were computed. The manually contoured prostate on rMRI and sMRI was compared using Dice-similarity coefficient (DSC). Interobserver variability was assessed using the DSC between two different radiation oncologist's rMRI prostate contours. Lastly, the dosimetric impact of sMRI was assessed using planning target volumes (PTV) generated from the sMRI and rMRI prostate contours.

Results: Line profiles illustrate enhanced prostate boundary visibility in sMRI compared with CT. sMRI pSNR was 69.2±1.4 dB and SSIM between sMRI and rMRI was 0.78±0.07 over the test datasets. The DSC between the rMRI and sMRI prostate contours was 0.851±0.047 vs. 0.837±0.070 (p=0.13) for the DSC between the rMRI prostate contours of two radiation oncologists. sMRI PTV V100% was 92.9%±3.8% vs. 93.1%±2.9% (p=0.40) for the rMRI PTV V100%.

Conclusion: This is the first report which includes the dosimetric impact of using sMRI in prostate HDR brachytherapy. Our results indicate the feasibility of MRI synthesis using a novel GAN framework, allowing a CT-only workflow while maintaining the MRI information.


Image-guided Therapy, HDR, MRI


TH- Brachytherapy: Imaging for brachytherapy: development and applications

Contact Email