Table 2.
Performance of SHG-based models for quantitative assessment of liver fibrosis in NAFLD
| Study | Model | Methodology | No. of patients | Performance |
|---|---|---|---|---|
| Wang et al. [79] (2017) | q-FP | SHG/TPEF to capture images (of whole biopsy section) | 50 (test cohort) | Principal component analysis model of 16 q-FPs: |
| Images assessed with computerized image-analysis by two independent investigators to output the profile of q-FPs data for each slice in operator-defined segmentation regions of liver tissue, including: | 42 (validation cohort) | - Fibrosis vs. no fibrosis: AUC 0.88 | ||
| (1) General: the liver section in its entirety | - Cirrhosis vs. earlier stages: AUC 0.93 | |||
| (2) Perisinusoidal: hepatocyte-associated collagen in the perisinusoidal space | Linear scale of fibrosis measurement of 4 q-FPs using desirability functions: | |||
| (3) Vessel: collagen fibrils directly connected to veins; and | - Related to fibrosis stage (P<0.0001) | |||
| (4) Vessel bridges: collagen fibrils extending from vein to vein or vein to portal tract. | ||||
| 70 q-FPs had interclass concordance ≥0.8 which were selected for further model development | ||||
| Wang et al. [81] (2020) | q-FP | Compared against NASH CRN staging system (but with substages of stage 1 combined) | 344 (428 biopsies) (larger validation study) | 25 q-FPs with AUC >0.90 for different fibrosis stages; perimeter of collagen fibres and number of long collagen fibres had the best accuracy (88.3–96.2% sensitivity and 78.1–91.1% specificity for different fibrosis stages) |
| Chang et al. [76] (2018) | SHG B-index | SHG/TPEF to capture images (final sampling size of 10 mm2 per biopsy) | 83 adults | Prediction model based on 14 unique SHG-based collagen parameters |
| An image processing algorithm was used to quantify fibrosis features in three specific regions: 1) central vein, 2) portal tract, and 3) perisinusoidal | - Fibrosis vs. no fibrosis: AUC 0.853 | |||
| In total, 100 collagen features were extracted and quantified, of which 28 features including the percentages of different collagen patterns and collagen string features were extracted in each region | - Cirrhosis vs. earlier stages: AUC 0.941 | |||
| - Stage 0/1 vs. 2/3/4: AUC 0.967 | ||||
| - Stage 0/1/2 vs. 3/4: AUC 0.985 | ||||
| Compared against Brunt’s staging system | - High correlation of 0.820 with fibrosis stage (P<0.001) | |||
| Liu et al. [77] (2017) | qFibrosis | SHG/TPEF to capture images (final sampling size of 10 mm2 per biopsy) | 62 adults (30 training, 32 validation); 36 children (18 training, 18 validation) | Prediction model based on six shared parameters for string collagen |
| An image processing algorithm was used to quantify fibrosis features in three specific regions: 1) central vein, 2) portal tract, and 3) perisinusoidal | (Adult) | |||
| - Fibrosis vs. no fibrosis: AUC 0.835 | ||||
| In total, 100 collagen features were extracted and quantified | - Cirrhosis vs. earlier stages: 0.982 | |||
| - Stage 0/1 vs. 2/3/4: AUC 0.892 | ||||
| - Stage 0/1/2 vs. 3/4: AUC 0.87 | ||||
| (Pediatric) | ||||
| - Fibrosis vs. no fibrosis: AUC 0.981 | ||||
| - Stage 0/1 vs. 2/3: AUC 0.931 | ||||
| - Stage 0/1/2 vs. 3: AUC 0.885 | ||||
| Liu et al. [82] (2020) | qFibrosis | Compared against NASH CRN staging system | 219 adults (146 training, 73 validation) (multicenter) | Prediction model based on 17 parameters, with output as a numerical index from 0 and 6.55 |
| - Fibrosis vs. no fibrosis: AUC 0.87 | ||||
| - Cirrhosis vs. earlier stages: 0.951 | ||||
| - Stage 0/1 vs. 2/3/4: AUC 0.881 | ||||
| - Stage 0/1/2 vs. 3/4: AUC 0.945 |
SHG, second harmonic generation; NAFLD, nonalcoholic fatty liver disease; q-FP, quantification of fibrosis-related parameter; TPEF, two-photon excitation fluorescence; AUC, area under curve; NASH CRN, Nonalcoholic Steatohepatitis Clinical Research Network.