抽象的

Discrete Tchebichef moment based machine learning method for the classification of disorder with ultrasound kidney images

S Bama, D Selvathi

Medical imaging provides a mean to visualize internal organs of human body, their anatomy as well as their functionality. Out of the many imaging modality such as CT, MRI, PET etc, ultrasound imaging of soft tissues is the safest one. However, the low contrast images of ultrasound modality poses the classification task as challenging one. A novel and effective approach for the classification of abnormalities in ultrasound B mode kidney images using Discrete Tchebichef Moment (DTM) is proposed. Instead of considering the whole kidney regions for the abnormality classification, typical blocks from the parenchyma (i.e cortex, medulla) and central sinus regions of kidney have been considered. Tchebichef kernels that are unique and not the rotational representation of existing kernels are carefully selected to extract the features from each of these blocks. A multiclass Support Vector machine (SVM) classifier with 5 fold cross validation has been used to carry out the classification. Performance of the proposed work is analysed with a dataset comprising of 160 samples from 40 images, Hu’s invariant features and Grey Level Co-occurance Matrix (GLCM) features are also extracted and fed as input to the multiclass SVM for comparative analysis. Quantitative evaluation metrics such as accuracy, precision, recall and fscore have been computed to reveal the performance of the proposed method. Promising results produced by the proposed work reveals the scope of Computer Aided classification of disorders in ultrasonic kidney images.