Prediction of Radiation-Induced Abnormality in Liver Enzymes from Machine Learning (ML) Algorithms

Abstract

Background: Exposure to ionizing radiation from medical radiation equipment during cancer diagnosis and treatment can alter the biochemistry of hospital personnel by triggering the oxidative stress process.

Aim: To develop a Simple-Linear-Regression algorithm with supervised learning applied to find the correlation between liver enzymes with the AAED (mSv) in low-dose medical radiation workers.

Methodology: Radiology & Nuclear Medicine Radiation workers from INMOLHospital were included. The AAED (annual average effective radiation doses) received from TLDs were measured by Radiation Dosimetry Laboratory. The models were trained and applied to the sample data set.

Results: The mean value of AAED was 0.28 mSv.  Half of the workers were found with high ALT levels and around 20% were found with altered AST levels. The models were also successfully cross-validated. ALT (R²= 0.025) & AST (R²= 0.00072) were having very weak relationships with AAED. From regression equations, it is inferred that for every unit increase in AAED (mSv), there will be a 12.98 unit decrease in ALT (U/L) and a 0.63 unit increase in AST (U/L) values.

Conclusion: Our ML model was successfully implemented to predict the alteration or abnormality in the liver enzymes from radiation exposure. It can assist physicians to detect changes in an individual's biochemistry before exposure to certain toxins.

Keywords: Radio-induced liver injury; Annual average effective radiation doses; Liver enzymes; Machine Learning (ML) Model