Computerised Analysis of Liver Texture with Correlation to Needle Biopsy

Zusammenfassung

Ziel: Verbesserung der Gewebecharakterisierung für die nichtinvasive Diagnose der diffusen Fettanreicherung in der Leber durch die Erstellung quantitativer Parameter für Ultraschall-Echos (US) in Korrelation mit der Histologie. Methoden und Material: Es wurden Ultraschall-Bilder von Patienten dokumentiert, die wegen anhaltend erhöhter Leberenzyme oder einer positiven Virushepatitis-Serologie zur Leberbiopsie (FNP) überwiesen wurden. Die histopathologischen Berichte wurden überprüft. Lebersteatose, Entzündung und Ausmaß der Fibrose wurden in Schweregrad 0 (normal) bis 3 (schwer) eingeteilt. Patienten mit einem Steatose-Grad von 3 ohne Zeichen der Entzündung oder Fibrose wurden ausgewählt. Ultraschall-Bilder von 24 Gesunden dienten als Kontrollen. Vier Gewebestruktur-Indices wurden für einen Zielbereich kalkuliert, der der Biopsie-Region entsprach. Sensitivität und Spezifität der Unterschiede zwischen den Gruppen wurden berechnet. Ergebnisse: Fettlebern und gesunde Lebern bildeten zwei klar unterscheidbare Gruppen. In allen Parameter-Unterbereichen gab es jedoch eine leichte Überschneidung zwischen den Gruppen, wobei einige Fälle jenseits der Dichotomie-Linie lagen. Es ergab sich eine hohe Sensitivität für alle Indices (90 - 100 %). Schlussfolgerung: Es ist möglich, eine höchst genaue „Ultraschall-Biopsie” der Fettleber zu erhalten. Die beschriebenen Indices können als Hilfsmittel für die computergestützte Ultraschall-Diagnose der diffusen Leberparenchymerkrankung Anwendung finden, insbesondere für die schwere Lebersteatose.

Abstract

Aim: To assist in tissue characterisation for the non-invasive diagnosis of diffuse fatty liver infiltration by providing quantitative indices of ultrasonic (US) backscatter with correlation to histology. Methods and Materials: US images from patients referred to US-guided liver needle biopsy (LNB) for persistently elevated liver enzymes or serologically positive markers for viral hepatitis were recorded. The histopathological reports were reviewed. Steatosis, inflammation and degree of fibrosis were scored from 0 (normal) to 3 (severe). Patients with level 3 steatosis without inflammation or fibrosis were selected. US images from twenty-four healthy subjects served as control. Four textural indices were calculated for a selected ROI corresponding to the biopsy site. Sensitivity and specificity of discrimination between the two groups were evaluated. Results: Fatty and healthy livers formed two distinct clusters. However, in all parametric subspaces there was a slight overlap between the groups with a few numbers of cases located across the dichotomy line.The sensitivity for all the indices was high (90 - 100 %). The specificity for each of the indices was moderate. The co-occurrence local homogeneity index yielded the highest specificity (88.5 %), with a sensitivity equivalent to two of the other indices (90 %). Conclusions: Highly accurate “ultrasonic biopsy” may be obtained for severe fatty liver. The described indices can serve as a tool in US computer- aided diagnosis (CAD) of diffuse parenchymal liver disease, in particular for severe steatosis of the liver.

Schlüsselwörter

Ultraschall - Gewebestruktur - Steatose - diffuse Leberparenchymerkrankung - Feinnadelbiopsie

Key words

Ultrasound - texture - steatosis - diffuse liver disease - needle biopsy

References

• 1 Sepulveda-Flores R N, Vera-Cabrera L, Flores-Gutierrez J P. et al . Obesity-related non-alcoholic steatohepatitis and TGG-beta1 serum levels in relation to morbid obesity.  Ann Hepatol. 2002;  1 36-39
• 2 Gangeh M J, Hanmandlu M, Bister M. A fuzzy-based texture analysis for tissue characterization of diffused liver diseases on B-scan images.  Biomedical Sciences Instrumentation. 2002;  38 369-374
• 3 Horng M H, Sun Y N, Lin X Z. Texture feature coding method for classification of liver sonography.  Comput Med Imaging Graph. 2002;  26 33-42
• 4 Mojsilovic A, Popovic M, Markovic S Krstic M. Characterization of visually similar diffuse diseases from B-scan liver images using nonseparable wavelet transform.  IEEE-Trans-Med-Imaging. 1998;  17 541-549
• 5 Sujana H, Swamamani S, Suresh S. Application of artificial neural networks for the classification of liver lesions by image texture parameters.  Ultrasound Med Biol. 1996;  22 1177-1181
• 6 Yao W, Zhao B, Zhao Y. et al . Ultrasonographic texture Analysis of parenchymatous organs by Four Neighborhood Pixels Algorithm.  J Ultrasound Med. 2001;  20 465-471
• 7 Dietrich C F, Wehrmann T, Zeuzem S. et al . Analysis of hepatic echo patterns in chronic hepatitis C.  Ultraschall Med. 1999;  20 (1) 9-14
• 8 Drescher T, Letterer H, Richter K P. Tissue characterization of the liver using ultrasound attenuation in an animal experiment.  Ultraschall Med. 1990;  11 (2) 104-108
• 9 Rath U, Zuna I, Limberg B. et al . Gray-level histogram analysis in the sonographic diagnosis of diffuse parenchymal liver disease.  Ultraschall Med. 1984;  5 (3) 94-97
• 10 Steinmaurer H J, Jirak P, Walchshofer J. et al . Accuracy of sonography in the diagnosis of diffuse liver parenchymal diseases - comparison of sonography and liver histology.  Ultraschall Med. 1984;  5 (3) 98-103
• 11 Haralick R M. Statistical and Structural Approaches to Texture.  Proceedings of the IEEE. 1979;  67 786-803
• 12 Gotlieb C C, Kreyszig H E. Texture Descriptors Based on Co-Occurrence Matrices.  Computer Vision, Graphics and Image Processing. 1990;  51 70-86
• 13 Bhanu P rakash KN, Ramakrishnan A G, Suresh S. et al . An investigation into the feasibility of fetal lung maturity prediction using statistical textural features.  Ultrasonic Imaging. 2001;  22 39-54
• 14 Tou J T, Gonzalez R C:. Pattern Recognition Principles. Massachusetts.  Addison Wesley Publishing Company. 1974; 
• 15 Matteoni C A, Younossi Z M, Gramlichi T. et al . Nonalcoholic fatty liver disease: a spectrum of clinical and pathological severity.  Gastroenteroly. 1999;  116 1413-1419
• 16 Abe C, Kahn C E Jr, Doi K. et al . Computer-aided detection of diffuse liver disease in ultrasound images.  Invest Radiol. 1992;  27 71-77
• 17 Layer G, Zuna I, Lorenz A. et al . Computerized ultrasound B-scan texture analysis of experimental diffuse parenchymal liver disease: correlation with histopathology and tissue composition.  J Clin Ultrasound. 1991;  12 192-210
• 18 Layer G, Zuna I, Lorenz A. et al . Computerized ultrasound B-scan texture analysis of experimental fatty liver disease: influence of total lipid content and fat deposit distribution.  Ultrason Imaging. 1990;  12 171-188
• 19 Oosterveld B J, Thijssen J M, Hartman P C. et al . Ultrasound attenuation and texture analysis of diffuse liver disease: methods and preliminary results.  Phys Med Biol. 1991;  36 1039-1064
• 20 Schuster E, Knoflach P, Grabner G. Local texture analysis: an approach to differentiating liver tissue objectively.  J Clin Ultrasound. 1988;  16 453-461
• 21 Gath I and Geva AB. Unsupervised optimal fuzzy clustering.  IEEE Trans Patten Anal Machine Intel. 1989;  11 (7) 773-781