Screening of hepatocellular carcinoma via machine learning based on atmospheric pressure glow discharge mass spectrometry.

Abstract

Hepatocellular carcinoma (HCC) is one of the most common malignant tumors with a high mortality rate. The diagnosis of HCC is currently based on alpha-fetoprotein detection, imaging examinations, and liver biopsy, which are expensive or invasive. Here, we developed a cost-effective, time-saving, and painless method for the screening of HCC via machine learning based on atmospheric pressure glow discharge mass spectrometry (APGD-MS). Ninety urine samples from HCC patients and healthy control (HC) participants were analyzed. The relative quantification data were utilized to train machine learning models. Neural network was chosen as the best classifier with a classification accuracy of 94%. Besides, the levels of eleven urinary carbonyl metabolites were found to be significantly different between HCC and HC, including glycolic acid, pyroglutamic acid, acetic acid, etc. The possible reasons for the regulation were tentatively proposed. This method realizes the screening of HCC via potential urine metabolic biomarkers based on APGD-MS, bringing a hopeful point-of-care diagnosis of HCC in a patient-friendly manner.