Shape-based boundary segmentation assessment for cervical vertebrae classification

Shape-based boundary representation is a critical component in medical imaging-based diagnosis system. Furthermore, accurate boundary representation is also essential in getting good segmentation, feature indexing and classification. This work concentrates on cervical vertebra distortion across the anterior boundary to detect anterior osteophytes and identify its severity. A suitable shape representation, which defines the vertebrae shape boundary must be determined prior to the cervical vertebra model can be constructed. Therefore, two different shape-based representations for vertebral segmentation using active shape model are analysed, which are 9-anatomical point representation (9-APR) and B-spline representation (BSR). The aims is to compare and quantify the precision and accuracy performance of the segmented vertebra with the help of orientation histogram-based indexing and multi-layer perception based on ten-fold cross validation (MLP-TFCV) classifier. Experimental results are assessed based on the segmented boundary outline by using ROC and AUC curves. The results show that the performance of both boundary representations are almost similar where B-SR has a small advantage compared to the 9-APR. The system can be further improved by introducing a statistical sampling method of the anterior points instead of the whole vertebra.