Artificial Intelligence for Replacing Manual Protein Abundance Assessment in Cirrhotic and Liver Cancer Tissue Samples

Abstract

Background: Hepatocellular carcinoma (HCC) is the primary form of liver cancer and patients with cirrhosis are at high risk of developing HCC. Loss of function of the transforming growth factor β (TGF-β) pathway is associated with HCC. Methods: Immunohistochemical analysis of biopsy tissue from patients with cirrhosis with or without HCC was performed to evaluate if the abundance of the receptor proteins TGFBR1 or TGFBR2 differentiated between HCC and cirrhosis. Samples were obtained from 3 independent sites. Analysis was performed using 3 separate sets of samples, and findings were validated by automated image analysis. Logistic regression modeling was performed to evaluate the predictive power of staining intensity of TGFBR1, TGFBR2, or the combination in differentiating HCC from cirrhosis. Findings: Analysis of a multi-institutional repository of tissue samples showed that the abundance of the TGF-β receptor protein TGFBR2 was significantly reduced in HCC tissue compared with cirrhotic tissue (p < 0.05). An artificial intelligence (AI)-based model that correctly identified cirrhotic and HCC tissue confirmed that TGFBR2 was significantly reduced in HCC tissue compared with cirrhotic tissue (p < 0.005). No single value of TGFBR2 staining intensity, with or without that of TGFBR1 staining intensity, effectively differentiated HCC from cirrhosis. Interpretation: TGFBR2 staining intensity alone is insufficient to diagnose HCC in cirrhotic tissue; however, a relative regional reduction in TGFBR2 in liver tissue samples of patients with cirrhosis may be a marker of HCC development. An automated image analysis pipeline could reduce variability in diagnosing HCC from biopsy tissue. Funding Information: This research work was supported by NIH grants R01AA023146 (L. Mishra), NIH R01CA236591 (L. Mishra), NIH U01 CA230690-01 (L. Mishra), VA Merit I01BX003732 (L. Mishra) and Elaine H. Snyder Cancer Research Award (L. Mishra). Declaration of Interests: These authors have no conflict of interest to declare. Ethics Approval Statement: The study was approved by relevant institutional review boards and all patients provided written informed consent before sample collection.