Poster Poster Program Therapy Physics

Synthetic MRI Derived from Treatment Planning CT In Head and Neck Cancer

Abstract
Purpose

To develop and rigorously assess a two‑stage deep learning framework capable of generating synthetic magnetic resonance images (sMRI) directly from standard planning computed tomography (CT) scans. The overarching aim of this work is to produce sMRI with soft‑tissue contrast and anatomical detail that closely mirrors true MRI, thereby addressing the long‑standing limitation of CT’s poor soft‑tissue visualization and enabling MRI‑like information to be incorporated into radiotherapy workflows even in settings where MRI access is limited.

Methods

This retrospective study included 60 patients with oropharyngeal cancer treated with radical radiotherapy between January 2023 and January 2024. Fifty patients were allocated to the training cohort for development of the CT‑to‑MRI deep learning model. Each of these patients underwent a standard planning non‑contrast CT scan as part of their radiotherapy workflow, followed immediately by a dedicated MRI examination on a Siemens 3T Magnetom Vida MR simulator. This ensured that high‑quality, temporally aligned CT and MRI datasets were available for deformable registration and subsequent model training. Deformable registration produced approximately 10,000 CT–MRI slice pairs for training. The proposed workflow consists of a two‑stage architecture: a Pix2Pix conditional GAN for initial sMRI synthesis, followed by a conditional diffusion model for refinement. For evaluation, 2,000 CT scans from an independent cohort of 10 test patients, each contributing both contrast and non‑contrast scans were used to generate sMRI. Model performance was assessed using standard image‑similarity metrics.

Results

For non‑contrast CT test cases, the model achieved an average MS‑SSIM of 0.82 ± 0.06. MS‑SSIM values for contrast‑enhanced CT were approximately 0.02 lower.

Conclusion

This study demonstrates that synthetic MRI generation from planning CT is feasible using a two‑stage deep learning approach. The generated images exhibit soft‑tissue contrast and grayscale patterns closely resembling real MRI.

People

Related

Similar sessions

Poster Poster Program
Jul 19 · 07:00
Python-Based Automation Framework for Annual Machine QA Data Archiving In Qatrack+

Annual water-tank measurements help ensure beam characteristics remain consistent with commissioning baselines. However, the lack of a standardized processing workflow and decentralized data storage makes it difficult to analyze...

Syed Bilal Ahmad, PhD
Therapy Physics 0 people interested
Poster Poster Program
Jul 19 · 07:00
User Expectations and Current Availability of HDR Brachytherapy Audits In Europe

The aim of this work was to evaluate the need to implement more dosimetric audits in high‐dose‐rate brachytherapy (HDR-BT) in Europe and to identify which characteristics such audits should meet according to users.

Javier Vijande, PhD Laura Oliver Cañamás
Therapy Physics 0 people interested