ABSTRACT
Background: The liver function tests (LFTs) remain one of the
most commonly employed clinical measures for the diagnosis of hepatobiliary
disease. LFTs sometimes referred to as hepatic panel help to determine the
health of liver, monitor the progression of a disease and measure the severity
of a disease particularly scarring or cirrhosis of the liver. Aims: In
this study, we present a new approach to evaluate the natural progression of
liver disease through the assessment of eight biochemical parameters:
serum total bilirubin (TB), alanine aminotransferase (ALT), aspartate
aminotransferase (AST), Alkaline phosphatase (ALP), total protein (TP), albumin
(ALB), albumin/globulin (A/G) ratio, and alpha-fetoprotein (AFP) as well as
two machine learning (ML) tools—Random Forest and CART to substantive the outcome. Methods: The study was carried out in a
total of 100 subjects which included healthy controls (group I-25 patients),
patients with acute hepatitis (group II-25 patients), chronic hepatitis (group
III-25 patients) and hepatocellular carcinoma (group IV-25 patients) applying
both biochemical and Machine Learning methods. Results: Of the eight
parameters tested, all except ALP (p = 0.426), showed an overall discriminatory
ability as judged by one-factor analysis of variance (p < 0.0001). We also
assessed the differences among group means by least significance difference
(LSD). The analysis showed that TB remained significantly elevated in groups
II, III, and IV as compared to controls (p < 0.05). ALP did not have any
discriminatory power among the four groups tested. ALT and AST were good
discriminators only between the control groups and groups II and III. TP, ALB,
and A/G ratio were decreased significantly in groups III and IV as compared to
controls. Group III and IV were almost indistinguishable using these biochemical
parameters except for AFP, which was found to be elevated only in group IV. The
accuracy of classification into different liver patient groups using random
Forest and CART was 94% and 95% respectively. Conclusion: Acute hepatitis
(group II) shows a higher level of AST, ALT and ALP compared to chronic
hepatitis (group III) and hepatocellular carcinoma (group IV). Two machine learning
algorithms also predicted and supported the same biochemical results by
correctly classifying liver disease patients. We also recommend that the AFP
test can be performed if hepatocellular carcinoma is suspected.