Abstract
Texture analysis can be used as a classification approach to describe microscopic changes in the liver. In our study, a total of 260 patients aged 4 to 90 underwent successful liver ultrasound examinations using a General Electric ultrasound machine (21045-87) with a 3.5MHz curve-linear transducer, typically used to scan the liver. The liver was scanned in multiple planes (transverse, sagittal, and oblique) to analyze the lesion based on shape, position, size, and echogenicity. Then the pictures were retrieved and classified into 5 categories: normal, a liver cyst, a hydatid cyst, hepatocellular carcinoma (HCC), and liver metastases. All pictures were 512 x512 pixels with 8-bit gray-level and were encoded in DICOM format; then three FOS features (mean, entropy, and energy, obtained from the intensity function of the images) were calculated for each ROI through all images using a 3x3 window size, and the data were processed for stepwise linear discriminant (SW-LD) analysis. The classification matrix of the original and predicted groups, using the discriminant function, presents the classification accuracy of each class in which 99.2% of normal liver was correctly classified and 75.6%, 81.4%, 100.0%, and 100.0% classification sensitivity for liver cyst, HCC, hydatid cyst, and liver metastases, respectively, with the highest predictive overall accuracy of 89.1%.(International Journal of Biomedicine. 2023;13(1):111-114.)