Towards thalassemia detection using optoelectronic measurements assisted with machine-learning algorithms: a non-invasive, pain-free and blood - free approach towards diagnostics

Abstract

In recent years, the convergence of electronic technology and machine learning has revolutionized healthcare sector which is poised to transform the field of clinical diagnostics. Enhanced accuracy, efficiency, and patient-friendliness for blood parameter estimations could be a boon to the mankind. Through this work we present the optoelectronically, photoplethysmography derived patient’s data for estimation of the hemoglobin levels using a non-invasive technique. The derived data is in good agreement with original complete blood count reports of individuals. Further over 800 complete blood count data obtained from pathological laboratories of various individuals were taken and subjected to machine learning algorithms, with input parameter as hemoglobin to predict hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, erythrocytes count and mean corpuscular hemoglobin concentration with good precision and accuracy. The predictions were correlated with the complete blood count data to estimate the normocytic anemia among patients. Comprehensive analysis, a series of statistical metrics, including mean absolute error, correlation coefficients, and R-squared scores have been used to obtain good accuracy to detect normocytic condition of a patient. Machine learning assisted, non-invasive technology for clinical diagnostics, especially in a very needy zone of thalassemia detection is hence proposed.