Second Harmonic Generation Reveals Subtle Fibrosis Differences in Adult and Pediatric Nonalcoholic Fatty Liver Disease

Feng Liu, MD; Jing-Min Zhao, MD; Hui-Ying Rao, MD; Wei-Miao Yu, MD; Wei Zhang, MD; Neil D. Theise, MD; Aileen Wee, MBBS, FRCPath, FRCPA; and Lai Wei, MD


Am J Clin Pathol. 2017;148(6):502-512. 

In This Article

Materials and Methods

Patients and Tissue Preparation

From 2011 to 2015, a total of 98 patients with consecutive biopsy-proven NAFLD from two centers (Peking University People's Hospital and 302 Hospital), comprising 62 adult patients (≥18 years old) and 36 pediatric patients (<18 years old), were included in this study. Patients with liver disease of other etiology, such as alcoholic or drug-induced liver disease, autoimmune liver disease, viral hepatitis, cholestasis, or genetic liver disease were excluded. This study was approved by the Ethical Committee of Human Experimentation in Peking University People's Hospital and 302 Hospital and was performed in agreement with the Helsinki Declaration of 1975. Patients were enrolled after providing written informed consent. LBx were serially sectioned for SHG imaging and stained with H&E, Masson trichrome, and Sirius red for histologic assessment. Clinicopathologic characteristics were obtained from the patients' medical records. Liver fibrosis was staged independently by two pathologists following the NASH Clinical Research Network (CRN) Scoring System,[8] and then discussed before reaching the final consensus. Patterns of steatosis were assessed as described in the NASH CRN study: azonal vs zone 1 vs zone 3 vs panacinar distributions. The fibrosis scores are as follows: 0, none; 1A, mild (delicate) zone 3 perisinusoidal fibrosis, requires Masson trichrome stain to identify; 1B, moderate (dense) zone 3 perisinusoidal fibrosis; 1C, portal fibrosis only; 2, zone 3 perisinusoidal fibrosis with periportal fibrosis; 3, bridging fibrosis; and 4, cirrhosis. The minimum length of liver biopsy specimens was 10 mm.

Image Acquisition System

Unstained sections of the tissue samples were imaged using a Genesis (HistoIndex, Singapore) system, in which SHG microscopy was used to visualize collagen, and the other cell structures were visualized with TPEF microscopy.

The samples were laser-excited at 780 nm, SHG signals were recorded at 390 nm, and TPEF signals were recorded at 550 nm. Images were acquired at ×20 with 512 × 512-pixel resolution, and each image had a dimension of 200 × 200 μm.[2] Multiple adjacent images were captured to encompass large areas. To cover most of the sample areas, 10 five-by-five multitile images were acquired for each biopsy sample with a final image size of 10 mm[2] (10 × 1 × 1 mm). For comparison, digitized images of liver tissue sections stained by H&E, Masson trichrome, and Sirius red were acquired using light microscopy (B X60; Olympus, Tokyo, Japan).

Image Quantification

One hundred collagen architectural features, including collagen percentages, specific collagen strings, and collagen connectivity-related measurements, were quantified from SHG/TPEF images to assess correlations with the NASH CRN fibrosis stages.

The overall collagen in SHG/TPEF images scanned from the liver tissue sections was divided into three specific regions: CV collagen (CV expansion), PT collagen (portal expansion and bridging fibrosis), and perisinusoidal collagen in the parenchyma. Besides these geographic distributions, collagen measurements were further performed for two different patterns—namely, distributed collagen (fine collagen) and aggregated collagen (large patches). For each pattern, collagen strings were extracted from all regions of collagen and characterized according to length and thickness (long vs short, thick vs thin). Detailed descriptions of all 100 parameters are given in the supplementary materials (all supplemental materials can be found at American Journal of Clinical Pathology online).

Statistical Analysis and Model Construction

Continuous variables are expressed as mean ± SD. Categorical variables are summarized by counts and percentages. Comparison of variables was performed using t tests for continuous variables and χ2 tests for categorical variables. Statistical significance was considered when P < .05. Spearman's rank correlation and its statistical significance were tested between clinical characteristics and collagen architectural features; the mean and standard error of mean were calculated for each NASH CRN stage for both patient groups.[8] The Wilcoxon rank-sum test was performed to estimate differences in collagen architectural features between different fibrosis stages, as well as between adult and pediatric patients.

For both adult (30 training and 32 validation) and pediatric (18 training and 18 validation) groups, multiple collagen architectural features were used to assess the NASH CRN stages, using a prediction model that was learned with the support vector machine method. The area under curve (AUC) was calculated based on the receiver operating characteristic, and it was used to measure the diagnostic accuracy of the prediction model with the application of NASH CRN stages (S0-S4). Four AUCs were calculated in differentiating (1) patients with S0 and those with S1, S2, S3, and S4; (2) patients with S0 and S1 and those with S2, S3, and S4; (3) patients with S0, S1, and S2 and those with S3 and S4; and (4) patients with S0, S1, S2, and S3 and those with S4, respectively.