MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Automated Air Region Delineation On MRI for Synthetic CT Creation

R Thapa*, E Ahunbay , H Nasief , X Chen, X Li , Froedtert Hospital and the Medical College of Wisconsin, Milwaukee, WI

Presentations

(Sunday, 7/29/2018) 4:00 PM - 4:55 PM

Room: Karl Dean Ballroom C

Purpose: Automatically separating air from other low-signal regions (especially bone, liver, etc.) is problematic due to noise/artifacts on MRI-images, resulting in errors in synthetic CT (sCT) generation for MRI-based radiation-therapy (RT) planning. This work aims to develop techniques to accurately and automatically determine air-regions in MRI.

Methods: CT and MRI (T2)-scans of phantoms with fabricated air-cavities and 14 abdominal cancer patients were used. From the phantom data, air-tissue boundaries in MRI were identified by CT-MRI registration. Two formulas relating the MRI-intensities of air (outside the phantoms/patients) and surrounding materials were established to auto-threshold air-regions. These formulas were refined based on patient MRI-data. The air containing regions in the abdomen were transferred from reference CT to daily MRI and expanded by 1cm to give a region to use thresholding. Expansion was necessary to account for possible inaccuracies in deformable image registration. The air-region determined inside this target region is further refined by texture based method. Twenty-three quantitative image texture features from the gray level co-occurrence matrix were extracted from auto-thresholded-regions. A naïve Bayesian classifier was trained using the extracted with leave-one-out cross validation technique to differentiate air from non-air voxels.

Results: Air-tissue boundaries in phantoms can be identified on MRIs within 1mm accuracy as compared to those on the CTs using the auto-thresholding method. The air cavity delineation was improved and validated using the MRI texture differences between air and tissues, as judged by the area under the ROC curve of 81% when two texture features (autocorrelation and contrast) were used. The performance increased to 82% with using three features (autocorrelation, sumvariance, and contrast).

Conclusion: The proposed techniques consisting of intensity-based auto-thresholding and image texture based voxel classification can automatically and accurately segment air-regions on MRI, allowing sCT to be generated quickly and precisely for MRI-based RT planning, especially for online-adaptation.

Keywords

Image-guided Therapy, MRI, Decision Theory

Taxonomy

IM/TH- MRI in Radiation Therapy: MRI/Linear accelerator combined (general)

Contact Email