Multiregional Soft Segmentation Driven by Modified ABC Algorithm and Completed by Spatial Aggregation: Volumetric, Spatial Modelling and Features Extraction of Articular Cartilage Early Loss

Abstract

In a clinical practise of the orthopaedics and medical imaging systems, the early cartilage loss, and cartilage lesions are challenging tasks. Due to an insufficient contrast, such pathologies are badly observable by naked eyes. Furthermore, objectification and quantification of those pathological findings are usually only subjectively estimated without the SW support. We propose a multiregional segmentation model based on the histogram classification with using of a sequence of triangular fuzzy functions where each such function represents specific knee area. To ensure a robustness of the model, respective fuzzy class location is driven by the ABC (Artificial Bee Colony) genetic algorithm respecting statistical features of the physiological cartilage. In the second step of the algorithm, a spatial aggregation is applied in order to consider spatial relationships in every region to prevent the image noise deterioration. Such multiregional segmentation model allows for an extraction of significant features well corresponding with the early cartilage loss like is the cartilage volume.