Time series analysis of the Twinkling Artifact in color Doppler sonography for surface roughness differentiation: An in vitro feasibility study

Color Doppler Twinkling Artifact (TA) images acquired from the internal body stones contain coded information about the roughness level of the tissue calculi which can be used for treatment management. The TA time series however have never been mathematically studied for roughness identification. This paper investigates the feasibility of estimating the roughness level of a surface by analyzing its TA time series. The TA data of a roughness phantom was used in this study in 2 classes and 1000 TA time series were extracted for each of the classes. Then, three subsets of temporal, spectral, and wavelet features were extracted from each time series. Next, the Bayesian and Support Vector Machines (SVM) classifiers were employed for roughness differentiations. The performance of the proposed method was investigated for cross-comparison of feature subsets, classifiers, and dimension reduction efficiency. Results showed that with only first two principle components projected from the extracted features, an accuracy of 96.06% was obtained which proves the feasibility of roughness recognition by time series analysis of the TA data.