The deep learning method based on convolutional neural network (CNN) was used to extract the relevant features of liver fibrosis classification from the multi-modal information of digital pathological sections, clinical parameters and biomarkers of a large number of existing cases of liver puncture, and the U-Net architecture of CNN was used to segment and extract the features of clinical medical images.

Patients with chronic hepatitis B underwent B-ultrasound-guided liver biopsy, and were divided into mild liver fibrosis group (fibrosis grade 0-1, S1), significant liver fibrosis group (fibrosis grade 2, S2), advanced liver fibrosis group and early cirrhosis group (fibrosis grade 3-4, S3-4) according to the pathological results.In this study, 200 patients with different degrees of liver fibrosis and 200 normal volunteers were collected from 2018 to 2022, and their clinical biochemical data, imaging data and peripheral blood samples were collected.The pathological microenvironment characteristics, imaging characteristics, clinical parameter characteristics and other data of patients were extracted, and the distillation learning method based on teacher-student model was adopted to develop and construct a multi-modal big data analysis model for accurate grading of liver fibrosis, so as to achieve a non-invasive intelligent grading diagnosis system for liver fibrosis.

Inclusion Criteria:

1. Age of 18-60 years old
2. The diagnosis of chronic hepatitis B is in line with the diagnostic criteria of China's 2019 Chronic Hepatitis B Prevention and Treatment Guidelines, and the diagnosis of non-alcoholic fatty liver is in line with the Asian Pacific Hepatology Association guidelines
3. Imaging showed no liver cancer

Exclusion Criteria:

1. There are contraindications for liver biopsy
2. Liver pathology did not meet the criteria

A patient with chronic hepatitis B liver fibrosis confirmed by liver biopsy