Premium
Hepatic steatosis estimated microscopically versus digital image analysis
Author(s) -
Hall Andrew R.,
Dhillon Amar P.,
Green Anna C.,
Ferrell Linda,
Crawford James M.,
Alves Venancio,
Balabaud Charles,
Bhathal Prithi,
BioulacSage Paulette,
Guido Maria,
Hytiroglou Prodromos,
Nakanuma Yasuni,
Paradis Valerie,
Quaglia Alberto,
Snover Dale,
Theise Neil,
Thung Swan,
Tsui Wilson,
Leeuwen Dirk J.
Publication year - 2013
Publication title -
liver international
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.873
H-Index - 110
eISSN - 1478-3231
pISSN - 1478-3223
DOI - 10.1111/liv.12162
Subject(s) - steatosis , digital image analysis , fatty liver , medicine , image (mathematics) , gastroenterology , computer science , artificial intelligence , computer vision , disease
Background & Aims Evaluate in liver biopsies: (i) interobserver agreement of estimates of fat proportionate area ( eFPA ) and steatosis grading, (ii) the relationship between steatosis grades and measured fat proportionate area ( mFPA , digital image analysis), (iii) the accuracy of eFPA , (iv) to present images to aid standardization and accuracy of eFPA . Methods Twenty‐one liver biopsies were selected from the Royal Free Hospital (RFH) histopathology archive to represent the full range of histopathological steatosis severity. As many non‐overlapping fields of parenchyma as possible were photographed at ×20 objective magnification from the biopsies ( n = 651). A total of 15 sample images were selected to represent the range of steatosis seen. Twelve hepatopathologists from 11 sites worldwide independently evaluated the sample images for steatosis grade [normal (none)/mild/moderate/severe], and eFPA (% area of liver parenchyma occupied by fat). Results The hepatopathologists had good linear correlation between eFPA and mFPA for sample images ( r = 0.924, P < .001) and excellent concordance (kappa = 0.91, P < 0.001). Interobserver concordance of steatosis grade showed ‘substantial agreement’ (kappa = 0.64). There was significant difference between eFPA and mFPA in the sample images for mild, moderate and severe steatosis ( P = 0.024, P < 0.001, P < 0.001 respectively): the observers consistently over‐estimated the eFPA . Conclusion Hepatopathologists showed ‘excellent’ interobserver agreement in eFPA and ‘substantial’ agreement in assigning steatosis grade (precision was high). However, compared with mFPA , eFPA was inaccurate. eFPA systematically exceeds mFPA ; generally the overestimation increases with severity of steatosis. Considering that non‐invasive technologies for estimating liver fat utilize histopathology as reference, such assessments would benefit from quantitative validation of visually estimated microscopic liver fat percentages.