Performance of Deep Learning Synthetic CTs for MR-Only Brain Radiation Therapy

Document Type

Conference Proceeding

Publication Date


Publication Title

J Med Phys


Purpose: Deep learning offers strong potential for accurate and rapid generation of synthetic CT (synCT) from MRI. However, synCT performance for dosimetric and image-guided radiation therapy (IGRT) endpoints needs to be carefully assessed. This work evaluates the performance of a novel generative adversarial network (GAN) synCT to support MR-only brain workflows. Methods: SynCT images for 11 brain cancer patients (6 radiosurgery, 5 partial brain) were generated from T1-weighted post-gadolinium MR images by applying a GAN model with a residual network (ResNet) generator and a convolutional neural network with 5 convolutional layers as the discriminator that classified input images as real or synthetic. Following rigid registration, clinical structures and plans derived from simulation CT (simCT) images were transferred to synCT images. Dose was recalculated on 13 simCT/ synCT pairs using fixed monitor units. 2D gamma analysis (2%/2 mm, 3%/ 1 mm) was performed to compare dose distributions at isocenter. Dose metrics (D(95%), D(99%) and D(0.035 cc) were assessed for the target and organ at risks (OARs). IGRT performance was evaluated via cone-beam CT to synCT/simCT registrations. For 6 patients, a semi-automated registration program rigidly registered planar KV images to synCT and CT digital reconstructed radiographs (DRRs). Results: Average 2D gamma analysis passing rates at 3%1 mm and 2%2 mm were 94.7% and 97.0%, respectively. Excellent agreement in dose metrics (D(95%), D(99%) and D(0.035 cc)) were observed (mean difference





First Page


This document is currently not available here.