Click here to


Are you sure ?

Yes, do it No, cancel

A Learning-Based MRI Classification for MRI-Only Radiotherapy Treatment Planning

Y Lei , H Shu , S Tian , T Wang , A Dhabaan , T Liu , H Shim , H Mao , W Curran , X Yang*, Emory University, Atlanta, GA


(Tuesday, 7/31/2018) 1:15 PM - 1:45 PM

Room: Exhibit Hall | Forum 6

Purpose: The application of MRI significantly improves the accuracy of radiotherapy target delineation for many disease sites due to its superior soft-tissue contrast over CT. However, MR images do not contain the necessary electron density information for dose calculation. This study aims to develop a learning-based classification method to segment routine T1W MRI into air, soft tissue and bone to generate pseudo CT for potential MRI-only photon or proton radiotherapy treatment planning.

Methods: We propose to integrate patch-based anatomical signature and an iterative refinement strategy into a machine learning framework to iteratively segment MRI into air, soft tissue and bone. The proposed segmentation of MRI consists of a training stage and a segmentation stage. During the training stage, patch-based anatomical features were extracted from the co-registered MRI-CT training images, and the most informative features were identified to train a classification forest with iterative refinement. During the segmentation stage, we extracted the selected features from the new MRI and fed them into the well-trained forests for MRI segmentation. The corresponding CT Hounsfield units (HU) can be assigned to three segmented masks (air, soft tissue and bone) for generating the MRI-based pseudo CT.

Results: This segmentation technique was validated with a clinical study of 11 patients with T1W MRI and CT images of the brain. Our classified results were compared with reference (planning) CT to quantitatively evaluate segmentation accuracy using Dice similarity coefficient (DSC). Overall, the mean DSCs for air, soft-tissue and bone segmentations were 97.79±0.76%, 93.32±2.35% and 84.49±5.50%, respectively.

Conclusion: We developed a novel learning-based approach to segment routine T1W MRI for pseudo CT generation required by MRI-only photon or proton radiotherapy treatment planning and demonstrated its reliability. This technique is a useful tool that could increase the feasibility of MRI-only radiotherapy workflow or PET attenuation correction of a PET/MRI scanner.


Not Applicable / None Entered.


IM/TH- image segmentation: MRI

Contact Email