Robust feature for transcranial sonography image classification using rotation-invariant Gabor filter

Abstract

Transcranial sonography is a new tool for the diagnosis of Parkinson’s disease according to a distinct hyperechogenic pattern in the substantia nigra region. In order to reduce the influence of the image properties from different settings of ultrasound machine, we propose a robust feature extraction method using rotation-invariant Gabor filter bank. Except the general Gabor features, such as mean and standard deviation, we suggest to use the entropy of the filtered images for the TCS images classification. The performance of the Gabor features is evaluated by a feature selection method with the objective function of support vector machine classifier. The results show that the rotation-invariant Gabor filter is better than the conventional one, and the entropy is invariant to the intensity and the contrast changes.