Modelování a extrakce oblasti artikulární chrupavky

Abstract

This dissertation thesis deals with the modeling and extraction of the articular cartilage from the MR (magnetic resonance) image data. In an area of the articular cartilage clinical diagnostic, a precise localization of the articular cartilage is essential with regard to the pathological processes which are often badly evaluable due to their manifestation is non contrast in comparison with the physiological cartilage. The cartilage assessment is often affected by the subjective error, and it is depended on physician’s experience. This work is focused on a design of the multiregional segmentation methodology which is based on two-phase pixel classification. In the first step, a design of the brightness segmentation procedure is figured out approximating individual image regions by a sequence of the fuzzy triangular membership functions. A localization of those functions is driven by the ABC (Artificial Bee Colony) algorithm representing a genetic evolutionary process. The second part of the methodology deals with the local aggregation procedure taking into account spatial pixels relationships, and allows for a modification of the brightness classification membership function. A substantial benefit of such system is robustness against the noise pixels and artefacts which are presented in the MR image data. Output of the segmentation procedure is a regional model of the articular cartilage which reliably reflects the articular cartilage physiological area from locations of early cartilage loss. Such changes are badly detectable from the native MR data. The proposed method has been also tested on synthetic variable sources of the image noise.