Machine learning and statistical analysis for diagnosis of hematological diseases

Abstract

Blood is a biological fluid composed mainly of water, red blood cells and other components and it is a non-Newtonian fluid. Red blood cells play an important role in the rheological properties of the blood and are the main responsible for the shear thinning behaviour of blood. Some hematological diseases can change the geometrical shape of red blood cells and thus their viscosity. In this work we have computed the viscosity of different blood samples that were obtained with a microfluidic device and normalized the viscosities for hematocrit using statistical analysis tools. We have also used different machine learning methods as Logistic Regressions or Artificial Neural Networks (ANN) to predict if a sample of blood corresponds to healthy blood or to a blood with an hematological disease. We have obtained different performance for the different methods, some of them with very good results and an accuracy of 94% of correct prediction has been achieved with an ANN model