Computer assisted characterization of diffused liver disease using image texture analysis techniques on B-scan images

In this study we evaluate the accuracy of image texture analysis techniques in the characterization of ultrasonic liver images. The texture analysis techniques used are the Fractal Dimension Texture Analysis (FDTA), the Spatial Gray Level Dependence Matrices (SGLDM), the Gray Level Difference Statistics (GLDS), the Gray Level Run Length Statistics (RUNL), and First Order gray level Parameters (FOP). The algorithms mere applied on three sets of u/s liver images, fatty, cirrhosis, normal, (30 samples each) all histollogically proven. Textural features were extracted using regions of interest (ROI) of 32x32 pixels in size. Results were classified using the K-Nearest Neighbor (K-NN) classifier. The FDTA and SGLDM were able to characterize the three sets with accuracy of 80%, GLDS achieved 78.9% while RUNL and FOP both achieved 77.8% accuracy. Combination of GLDS and RUNL gave 81.1%, while combination of FDTA and SGLDM gave 82.2% accuracy.