Quantitative assessments of liver fibrosis using second-harmonic generation/two-photon excited fluorescence microscopy provide greater sensitivity and accuracy than collagen proportionate area while eliminating operator-dependent variation in the staining process. In conjunction with sophisticated image analysis algorithms and feature selection, we might reduce the computation cost in future and narrow down the candidates for further clinical studies.

We sampled a total of 244 liver specimens from patients with hepatitis B viral infections who underwent liver biopsy or liver resection at the National Taiwan University Hospital. The samples were then imaged using a Genesis (HistoIndex Pte. Ltd, Singapore) system, wherein second-harmonic generation microscopy was used to visualize collagen, and two-photon excited fluorescence microscopy was used to visualize other cell structures. We used 100 morphological features extracted from the images to assess correlations with METAVIR fibrosis scores.

Out of 100 quantitative measurements, 76 showed significant correlation with METAVIR scoring, thereby enabling the statistical discrimination of patients in various stages of the disease. These 76 features were also narrowed down by the nonlinear test to 10 candidate measurements, which can be further investigated in detail.

Our experimental results showed that the model with 10 selected features can beat the one with second-harmonic generation only, and performed equivalently well compared the model with 76 features, especially for early-stage discrimination. Features presenting significant correlation were used to fit a single combined index in order to predict pathological staging, thereby making it possible to reveal incremental progress during treatment.