Journal of Hepatology 2000; 32: 453–464 Copyright C European Association for the Study of the Liver 2000 Printed in Denmark ¡ All rights reserved Munksgaard ¡ Copenhagen Journal of Hepatology ISSN 0168-8278 Automatic quantification of liver fibrosis: design and validation of a new image analysis method: comparison with semi-quantitative indexes of fibrosis Marco Masseroli, Trinidad Caballero, Francisco O’Valle, Raimundo M.G. Del Moral, Alejandro Pe ´rez-Milena and Raimundo G. Del Moral Department of Pathology, School of Medicine and University Hospital, University of Granada, Granada, Spain Background/Aims: Liver fibrosis is one of the most important and characteristic histologic alterations in progressive and chronic liver diseases. Thus, in both clinical and experimental practice, it is fundamental to have a reliable and objective method for its precise quantification. Several semi-quantitative scoring sys- tems have been described. All are time-consuming and produce partially subjective fibrosis evaluations that are not very precise. This paper describes the design and validation of an original image analysis-based ap- plication, FibroQuant, for automatically and rapidly quantifying perisinusoidal, perivenular and portal- periportal and septal fibrosis and portal-periportal and septal morphology in liver histologic specimens. Methods: The implemented image-processing algo- rithms automatically segment interstitial fibrosis areas, while extraction of portal-periportal and septal region is carried out with an automatic algorithm and a simple interactive step. For validation, all automati- cally extracted areas were also manually segmented and quantified. F  is common in many progressive and chronic liver diseases and is one of the main histo- logic features considered for their diagnosis and prog- nosis (1). Therefore, its precise and objective histologic quantification is extremely important both in in vivo experimental models and in clinical practice. Until now, the histologic extent of liver fibrosis has been esti- mated mainly by conventional and modified semi- quantitative scoring systems (1–6). However, even if Received 21 April; revised 5 August; accepted 31 August 1999 Correspondence: Marco Masseroli, Dpto. de Anatomı ´a Patolo ´gica, Facultad de Medicina, Avda. de Madrid 11, 18012 Granada, Spain. Tel: 34 958 244097. Fax: 34 958 243510. e-mail: masseroli/biomed.polimi.it 453 Results: Statistical analysis showed significant intra- and interoperator variability in manual segmentation of all areas. Automatic quantifications did not signifi- cantly differ from mean manual evaluations of the same areas. Comparison of our image analysis quanti- fications with staging histologic evaluations of liver fibrosis showed significant correlations (Spearman’s, 0.72∞r∞0.83; p∞0.0001) and that the latter are based more on the distribution patterns than on the quantity of fibrosis. Conclusions: FibroQuant is a sensitive, precise, objec- tive and reproducible method of fibrosis quantifi- cation, which complements semi-quantitative histo- logic evaluation systems. This novel tool could be of special value in clinical trials and for improving the prognosis and follow-up among patients with fibrosis- inducing hepatic diseases. Key words: Automatic quantification; Image pro- cessing; Liver fibrosis; Portal-periportal morpho- metry. these semi-quantitative methods describe well the pathologic patterns of the hepatic structure, they pro- duce fibrosis evaluations that are not very precise and, especially at intermediate grades, are subjectively de- pendent on the visual interpretation of the observer, who must be an experienced pathologist (5–7). In recent years, quantitative evaluations through im- age analysis have been utilized (6,8–10). Digital image analysis rapidly provides objective quantitative results similar to but more precise than those determined by semi-quantitative scoring methods, without requiring the presence of an experienced pathologist (11). How- ever, due to the lack of specific automatic applications, the use of image analysis has been limited to quantifi- cation on the entire image of areas densitometrically