MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Patient-Specific Pseudo-CT Generation Using Semantic and Contextual Information-Based Random Forest for MRI-Only Radiation Treatment Planning

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

Presentations

(Thursday, 8/2/2018) 1:00 PM - 3:00 PM

Room: Karl Dean Ballroom B1

Purpose: MRI is increasingly being used in radiotherapy planning owing to its superior soft-tissue contrast without ionizing radiation compared with CT. However, MRI data do not contain electron density information that is necessary for accurate dose calculation. The purpose of this work is to develop a learning-based method generate patient-specific pseudo-CT (PCT) from routine anatomical MRI for MRI-only radiotherapy treatment planning.

Methods: We propose to integrate semantic and contextual information and patch-based anatomical signature into machine learning framework to iteratively predict CT from MRI based on auto-context model (ACM). The proposed CT prediction consists of the training stage and the prediction stage. During the training stage, patch-based anatomical features are used to sequentially train a series of classification random forests which create semantic and contextual information (discriminative probability maps). Finally, the features with these information are used to train a sequence of regression forests based on ACM. During the prediction stage, we extract the anatomical features from the new MRI and feed them into the well-trained classification and regression forests for PCT prediction and iterative refinement. Our PCT were compared with the planning CT to quantitatively evaluate the prediction accuracy.

Results: This prediction technique was clinically validated using 11 patients with head MRI and CT. The mean absolute error (MAE), peak signal-to-noise ratio (PSNR), normalized cross-correlation (NCC) and Dice indexes were used to quantify the prediction accuracy. Overall the MAE, PSNR and NCC were 27.30±5.06 HU, 31.14±1.75 dB and 0.97±0.01 for 11 patients’ data, and the Dice coefficients for 11 patients’ air, soft-tissue and bone were 97.79±0.76%, 93.32±2.35% and 84.49±5.50%.

Conclusion: We have investigated a novel learning-based approach to predict CT from routine MRI and demonstrated its reliability. This CT prediction technique could be a useful tool for MRI-based radiation treatment planning and MRI-based PET attenuation correction of PET/MRI scanner.

Keywords

Not Applicable / None Entered.

Taxonomy

IM/TH- MRI in Radiation Therapy: Development (new technology and techniques)

Contact Email