A Pilot Study for the Prediction of Liver Function Related Scores Using Breath Biomarkers and Machine Learning
Abstract Liver function test is the first step to diagnose various liver diseases by measuring certain proteins and liver enzymes from the blood sample. After getting the required data of the clinical parameters from the blood test it is possible to calculate Child-Pugh (CTP), AST to PLT ratio (APRI) and Model for end-stage liver disease (MELD) clinical scores that help the doctors about the severity of the disease progression. Volatile organic compounds (VOCs) found in-breath and monitoring their concentration may be a prevailing method for disease diagnosis. In this work, Isoprene, Limonene, and Dimethyl sulphide (DMS) are considered as a potential breath biomarker related to liver disease. A dataset is designed, that includes the biomarkers concentration analysed from the breath sample before and after study subjects performed an exercise. Four regression methods are performed to predict the clinical scores using breath biomarkers data as features set by the machine learning techniques. Regression methods on the dataset for prediction are evaluated by mean absolute error and root mean square error. A significant difference observed for isoprene concentration (p<0.01) and for DMS concentration (p<0.0001) between liver patients and healthy subjects breath sample. Ensemble regression methods are found best suited for the dataset. The mean absolute error for CTP score is 0.07, for APRI score 0.1 and for MELD score 0.7. The R-square value between actual clinical score and predicted clinical score is found to be 0.78, 0.95, and 0.85 for CTP score, APRI score, and MELD score, respectively. These results suggest that breath biomarkers hold a promising approach for non-invasive test and mass screening related to liver disease.