Identifying Possible Biomarkers for Early-Stage Hepatocellular Carcinoma using Random Forest Machine Learning Method

Abstract

Hepatocellular carcinoma is a primary liver tumour arising from hepatocytes, the liver's own cells. It is one of the most common types of cancer in the world. The most important cause is chronic liver disease due to hepatitis B and C infections. In some patients, HCC causes symptoms such as abdominal pain, loss of appetite, anaemia, nausea, fatigue and jaundice and is diagnosed as a result of tests. In some patients, it is detected incidentally by liver ultrasound, tomography or MRI performed for another reason. The most typical finding is an increase in a substance called alpha-fetoprotein (AFP). Although this does not occur in all patients, elevated AFP in a patient with cirrhosis strongly indicates the presence of HCC. HCC can be seen on ultrasound, tomography or MRI films. Especially in tomography and MRI, the rapid and strong retention of the intravenous drug and then its early wash out is a typical finding and if detected in a patient with cirrhosis, HCC can be diagnosed without the need for biopsy. However, in many patients, imaging findings are not typical and a biopsy is required for diagnosis. In this study, a Random Forest machine learning model was created with proteomic data regarding the cancerous tumor tissue and the adjacent non-cancerous tissue of 19 HCC patients. the accuracy, balanced accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1-Score, MCC and G-Mean values for the Random Forest model were 0.90, 0.88, 0.90, 0.93, 0.82, 0.91, 0.82 and 0.91, respectively. Considering the model-dependent variable significance, SRSF1 and PBLD proteins are suggested as biomarkers that may be clinically useful in the diagnosis of early-stage HCC.