Ion, Adina Izabella (2013) Computer aided detection and measurement of peripheral arterial diseases from CTA images. (PhD thesis), Kingston University, .
Abstract
Peripheral Arterial Disease (pAD) afflicts more than 2.7 million people in the U.K. per year, and it is projected to increase rapidly within the current decade. PAD is a product of obstruction (stenosis or occlusion) of vessels feeding the body's extremities, and it is most often encountered in the lower extremities. Treatment of the disease is dependent on the specific anatomic segments afflicted, the degree of stenosis and its length. A common technique for imaging PAD is Computed Tomography-Angiography (CTA). The acquired CTA images are then investigated by a radiologist for disease assessment. However due to the large size of the PAD CTA datasets (1000-2000 slices) the radiologist's examination is time consuming and laborious. This project brings a contribution to the investigation of PAD in CTA datasets by the development of a tool for the radiologist, a fully automatic system for the detection and measurement of PAD, as currently there are no such systems efficacious for the disease. The proposed system is comprised of two components: a Computer Aided Detection (CAD) component and a Computer Aided Measurement (CAM) component. The CAD component is designed for artery segmentation and stenosis detection. The stage of artery segmentation is accomplished by using a 3D region growing method and an innovative 3D fast morphology operation. CAD methodologies commonly employ morphological operations as a tool in the segmentation process, along with extended series of CTA images. This large dataset requires careful attention to be paid towards optimizing the computational process in terms of time efficiency. In order to meet this goal, an optimized morphology algorithm is presented, which reduces the computation time by a factor of 10. A skeletonization based centreline technique is applied on the detected artery, and it then provides the basis for the measurement stage. Orthogonal planes to the centreline are used in order to obtain cross sectional images. The artery profile is then built based on vessel areas measured in the cross sectional images and an automated process of stenosis detection is performed. The CAM component of the system accurately measures and quantifies the stenosis and overcomes the challenge brought by the partial volume effect. In this respect, a hybrid method for partial volume correction is employed locally, on the candidate areas of stenosis detected by the CAD component, based on Maximum a Posterior (MAP) and Markov Random Field (MRF) expectation maximization method. The CAD-CAM system has been successfully implemented and applied on phantom and patient data (twenty data sets from The University Hospital of Lausanne (CHUV)) and the evaluation was carried out through the visual judgment of two experienced radiologists. Within the CAD component, the artery segmentation was evaluated and a total of 15 peripheral arterial trees were correctly extracted. The proposed stenosis detection method was evaluated on 525 arterial segments (each dataset was partitioned into 35 segments) from which 132 exhibited stenosis caused by soft plaque. The system achieved a sensitivity of 88% and a specificity of 96%. The CAM component has been evaluated using phantom data, and the average error of the diameter measurement was 8%.
Actions (Repository Editors)
 |
Item Control Page |