Portal de Periódicos da CAPES

Local texture analysis: An approach to differentiating liver tissue objectively

1988; Wiley; Volume: 16; Issue: 7 Linguagem: Inglês

10.1002/jcu.1870160702

1097-0096

E. Schuster, P Knoflach, G. Gräβner,

Radiomics and Machine Learning in Medical Imaging

Abstract The usefulness of local textural analysis in identifying diseases of the liver has been evaluated. In comparing two areas texturally, a random distribution of one or several textural parameters indicates identical tissue, whereas a distinct clustering can be used to discriminate between different tissues. Using this procedure, various diseases of the liver were differentiated accurately whereas different images of the same state of the liver were classified as similar with high accuracy. The consistent results with different patients, sections, sample locations and scanner parameter settings demonstrate that these factors have no impact on the characteristics of the resulting pattern used for discrimination and therefore for the final diagnosis. The overall diagnostic accuracy of this method seems better than conventional sonographic evaluation of parenchymal liver disease and at least competitive with other computerized methods described. The most important result clinically seems to be the high accuracy in differentiating between hepatitis and normal liver, which cannot be achieved by conventional ultrasonography.