A deep-learning neural network model for blood platelet counting with the use of blood glucose, blood type, and other complete blood count parameters and physical properties in human adults

• Main Authors: Mariño, Alyssa Ysabelle, Encisa, Ronald Jeriko Ignacio
• Format: text
• Language: English
• Published: Animo Repository 2022
• Subjects: Neural networks (Computer science); Blood platelets; Physics
• Online Access: https://animorepository.dlsu.edu.ph/etdb_physics/22; https://animorepository.dlsu.edu.ph/context/etdb_physics/article/1013/viewcontent/2022_Encisa_Marino_A_deep_learning_neural_network_model_Full_text.pdf
• Institution: De La Salle University

An Artificial Neural Network (ANN) is a program that can apply to numerous methods of analyzing large amounts of data with the set goal of predicting an output given a set of inputs. In this study, ANN was utilized to examine two data sets, one coming from a training set composed of the Complete Blood Count Quality Control Values made available by the Centers for Disease Control (CDC). The other is the testing set in which the researchers gathered 32 subjects’ CBC test results, subdivided into two subgroups, diabetic and nondiabetic, as a means to present the correlation that was previously proven by the study of Akinsegun in 2014, wherein diabetics were discovered to have a higher mean platelet count than nondiabetics. Their blood types were also noted as another correlation presented by Okeke in 2020, wherein it was found that blood type O groups have been proven to have a lower platelet count than blood type A and B groups. The CBC parameters used as input for the ANN are as follows: basophils, eosinophils, hematocrit, hemoglobin, lymphocyte, mean cell hemoglobin concentration, mean cell hemoglobin, mean cell volume, monocyte, neutrophil, platelet count, red cell distribution width, red blood cells, and white blood cells. With 12 diabetic (D) and 20 nondiabetics (ND) subjects, the ANN was able to create a prediction model for platelet count based on the thirteen (13) CBC parameters of each subject. The indication of the model’s performance is based on the RMSE, the difference between true output and predicted output. It displayed the values 3.0203 and 4.2454 for the training and validation set, respectively, displaying a small error. The results of the study corresponded to the theory of Akinsegun, given that it showed a 91.67% of the twelve (12) diabetic samples had projected a higher mean platelet count than the predicted platelet count compared to the 50% of the twenty (20) nondiabetic samples from the ANN model. However, for the blood type group parameter, only two (2) out of (10) the blood type group O displayed a lower platelet count than the predicted value and had the lowest percent error of 1.65%; therefore, it did not correspond to the theory of Okeke in 2020.