Analysis method and system for early hepatocellular carcinoma postoperative recurrence prognosis based on artificial intelligence (AI)

Abstract

The invention relates to the technical field of prognosis intelligent algorithms, in particular to an analysis method and system for early hepatocellular carcinoma postoperative recurrence prognosis based on artificial intelligence (AI). The early hepatocellular carcinoma postoperative recurrence risk and prognosis are predicted through a complete digital hematoxylin-eosin staining (HE) staining tissue pathological section, and the method comprises the specific steps that a full-view digital section (WS I) image of a specimen to be analyzed is obtained; a foreground region of interest (ROI) is extracted from the WS I image; according to the invention, different areas of the digital pathological image are identified through deep learning. The early-stage hepatocellular carcinoma postoperative recurrence risk and prognosis are effectively analyzed directly according to the HE staining section, the accuracy is very high while multiple different cell regions are identified, and the invention has great guiding significance for patient recurrence.

Description

technical field [0001] The invention relates to the technical field of prognostic intelligent algorithms, in particular to an AI-based analysis method and system for postoperative recurrence and prognosis of early hepatocellular carcinoma. Background technique [0002] Liver cancer is the sixth most common malignant tumor in the world, among which hepatocellular carcinoma (Hepatocellular carcinoma, HCC) accounts for about 85% of primary liver cancer. Early liver cancer is the best indication for radical resection. The 5-year survival rate of patients with Barcelona Clinic Liver Cancer staging (BCLC staging) stage 0-A can reach 70%, but half of the patients suffer from recurrence. Therefore, overcoming the recurrence problem of early liver cancer is the top priority. At present, the recurrence prediction model of HCC involves tumor gross morphology, biochemical indicators, and genetic characteristics, and there are few explorations at the histological level. [0003] In re...

AI Technical Summary

Problems solved by technology

[0005] 1) The existing algorithm can only identify cancer areas, normal liver tissue, necrotic areas, and fibrous tissue, but does not identify important portal areas and lymphocytes in the tumor microenvironment, which will reduce the efficiency of the model

[0006] 2) The existing models focus on the prediction of the overall survival rate, and the prompting significance for the risk of postoperative recurrence is insufficient

[0007] 3) Existing algorithms fail to effectively normalize pathological slices from different institutions (such as for The CancerGenome Atlas, slice validation in the TCGA database), thus affecting model generalization

[0008] 4) The specificity and sensitivity of the existing models need to be improved

Images