Development of a Semi-Automatic Technique for Flow Estimation using Optical Flow Registration and k-means Clustering on Two Dimensional Cardiovascular Magnetic Resonance Flow Images

TY - CONF

T1 - Development of a Semi-Automatic Technique for Flow Estimation using Optical Flow Registration and k-means Clustering on Two Dimensional Cardiovascular Magnetic Resonance Flow Images

AU - Brix, Lau

AU - Christoffersen, Christian P. V.

AU - Kristiansen, Martin Søndergaard

AU - Sørensen, Thomas Sangild

PY - 2009

Y1 - 2009

N2 - Background: Manual analysis of cardiac magnetic resonance (CMR) flow data is a trivial but cumbersome task and must be carried out by an experienced radiologist on a dedicated workstation. Purpose: To construct a system that semi-automatically carries out a flow analysis on cardiac CMR data of the aorta. Methods: 2D phase contrast flow images of the aorta were acquired from a patient with an enlarged pulmonary artery on a Philips Achieva 1.5T CMR system. The cardiac motion was removed from the data set using the Cornelius/Kanade registration algorithm. The time resolved flow data was then categorized into groups by the k-means clustering method. Finally, the cluster containing the vessel under investigation was selected manually by a single mouse click. All calculations were performed on a Nvidia 8800 GTX graphics card using the Compute Unified Device Architecture (CUDA) extension to the C programming language. Results: Seven clusters were created and identification of the one including the aorta was hereafter trivial. However, a part of the rim of the aortic vessel was excluded from the main aortic cluster. Conclusion: The registration and clustering approach for analyzing CMR flow data seems promising because it saves time for post-processing. However, the k-means cluster approach is not comprehensive for quantitative flow estimations as it is but seems feasible for a subsequent segmentation algorithm like deformable contours (i.e. snakes). Future work may overcome this manual part and make the flow analysis completely automatic.

AB - Background: Manual analysis of cardiac magnetic resonance (CMR) flow data is a trivial but cumbersome task and must be carried out by an experienced radiologist on a dedicated workstation. Purpose: To construct a system that semi-automatically carries out a flow analysis on cardiac CMR data of the aorta. Methods: 2D phase contrast flow images of the aorta were acquired from a patient with an enlarged pulmonary artery on a Philips Achieva 1.5T CMR system. The cardiac motion was removed from the data set using the Cornelius/Kanade registration algorithm. The time resolved flow data was then categorized into groups by the k-means clustering method. Finally, the cluster containing the vessel under investigation was selected manually by a single mouse click. All calculations were performed on a Nvidia 8800 GTX graphics card using the Compute Unified Device Architecture (CUDA) extension to the C programming language. Results: Seven clusters were created and identification of the one including the aorta was hereafter trivial. However, a part of the rim of the aortic vessel was excluded from the main aortic cluster. Conclusion: The registration and clustering approach for analyzing CMR flow data seems promising because it saves time for post-processing. However, the k-means cluster approach is not comprehensive for quantitative flow estimations as it is but seems feasible for a subsequent segmentation algorithm like deformable contours (i.e. snakes). Future work may overcome this manual part and make the flow analysis completely automatic.

M3 - Poster

T2 - 27. Danske medicotekniske landsmøde

Y2 - 22 September 2009 through 24 September 2009

ER -