Shape analysis of healthy and diseased cardiac ventricles using PCA

Analysis of shape began to emerge as a useful method in medical image computing with the potential to improve the accuracy of medical diagnosis as well as the understanding of processes behind growth and disease. Our work aims at investigating the suitability of the medial representation method (m-reps) to model and analyze shape and shape differences between healthy and diseased hearts. For this experimental study we used MRI short axis scans of 16 male subjects, 11 healthy volunteers (age: 50 ± 10) and 5 patients (age: 57 ± 11) with dilatative cardiomyopathy (CMP, EF: 22 ± 10%). Controlled semi-automated segmentation using the level set method provided labels, which were used for the modeling process. To evaluate the model to image quality the similarity index (SI) was calculated. The SI for the left ventricle was 0.93, and for the right ventricle SI = 0.9. Applying extended Principal Component Analysis (PCA) on all 16 subjects revealed the distribution of the individual shapes, showing discrimination between healthy and diseased ventricles. The first two main components of the PCA, mainly responsible for discrimination, cover more than 40% of the whole shape space. Applying the medial based approach and using shape statistics using component analysis for the left and right ventricle is feasible. This preliminary work promises to discriminate at least globally based on shape differences between healthy and diseased hearts. The resulting question however is, if this approach is good enough to detect early structural changes leading to diseased hearts.