Evaluation of B-mode ultrasonographic images of carotid lesions by computer analysis as compared with visual...

To establish an objective method for evaluating the characteristics of carotid artery lesions by B-mode ultrasonography, we compared the reproducibility of computer analysis with that of visual evaluation. Using two different models of ultrasonograph, we evaluated B-mode ultrasonographic images of 100 carotid artery lesions in a blinded fashion by computer analysis and visual evaluation, two times at a 1-month interval. Another 35 lesions were simultaneously imaged by the two models of ultrasonograph and were evaluated visually and by computer analysis to examine the correlation between results. For visual evaluation, carotid artery lesions were divided into the following 3 types based on texture and relative echogenicity as compared with that of adjacent muscle tissue: heterogeneous, hypoechoic, and hyperechoic. For computer analysis, images were computerized by means of a scanner (using Adobe Photoshop software). The echogenicity of all pixels included in the lesion and adjacent areas of muscular tissue was measured, and the lesion was classified into the following 3 types based on the shape of histograms of echogenicity: mixed-echoic (M), hyperechoic (H), and lucent (L). The reproducibility of the results derived by computer analysis had a significantly higher coincidence coefficient (kappa = 0.89) than those derived by visual evaluation (kappa =0.68). The coincidence coefficient (kappa) of the same lesions depicted by two different models of equipment was 0.77 for computer analysis as compared with only 0.32 for visual evaluation. These findings indicate that computer analysis of histogram shape and lesion echogenicity can improve intra-sonographer and inter-equipment reproducibility in the evaluation of lesion echogenicity and texture.