Segmentation And Computer-aided Diagnosis Of Cardiac MR Images Using 4-D Active Appearance Models

The four-dimensional (4-D) cardiac MR images contain rich information about the static and dynamic properties of the heart, which were not fully utilized in clinical practice for quantitative analysis -- a difficult task for humans, which can be achieved by computer-aided image analysis and diagnosis. In this thesis, the 4-D Active Appearance Model (AAM) was used to achieve highly automated computer segmentation of the left and right ventricles (LV and RV) and the diagnosis of normal and tetralogy of Fallot (TOF) patients. The whole process was implemented in four stages: data construction, model construction, computer segmentation, and computer-aided diagnosis.

The data construction stage overcame most inherent limitations of cardiac MR imaging and produced high-quality 4-D ventricular image with isotropic voxels, complete coverage and no respiratory motion artifacts. A manual tracing application was developed to trace the ventricular surfaces in a true 4-D context and produced accurate independent standard for model construction and segmentation validation.

In the model construction stage, the 4-D AAMs were constructed using a custom designed automatic landmarking and texture mapping procedure with high efficiency.

In the computer segmentation stage, the 4-D AAMs were applied to segment the left and right ventricles of 25 normal and 25 TOF patient scans. The segmentation achieved accurate results measured by signed surface positioning errors. On normal hearts, the average signed errors were 0.3±2.3 mm for LV and 0.1±3.4 mm for RV. On TOF hearts with large shape variability, the errors were -1.5±3.2 mm for LV and -0.9±4.3 mm for RV. Other error metrics such as relative overlapping also indicated good segmentation accuracies.

In the computer-aided diagnosis stage, 100% normal/TOF classification was achieved using the novel 4-D ventricular function indices -- the shape modal indices. The longitudinal analysis performed on subjects with multiple annual scans showed that the normal subjects exhibited smaller variances of these 4-D indices than TOF patients, which demonstrated the potential of using them as disease status determinants. In addition, the quantitative 4-D indices provided more information about the dynamic properties of the heart and identified patient-specific features that were not sensed by human expert observers.