Performance comparison of artificial intelligence techniques in the diagnosis of hematological diseases

The most common hematological diseases are anemia and anemia-related diseases. Anemia, defined by the World Health Organization as a condition in which red blood cells or their oxygen-carrying capacity are insufficient, especially affects women and preschool children. Since anemia, significantly reduces the quality of life, is both a disease and a symptom accompanying serious diseases, its treatment can be vital. An accurate diagnosis process is also required for effective treatment. The increasing number of patients and the congestion in hospitals make it difficult for patients to reach expert medical doctors. Because of these difficulties, a system to recognize anemia in general practice conditions would be useful. Using this system with the first examinations requested for diagnosis in the primary health care services are provided, non-specialist personnel working in these health centers will provide more effective guidance. This study aims to perform the diagnosis of anemia and anemia-related diseases with artificial learning methods. For this study, an artificial learning architecture that can diagnose 12 different types of anemia developed. With the help of an interface first, completely original data without any numerical intervention obtained. The data are taken from Düzce University Research and Application Hospital with the permission of the ethics committee. At the next step, the weights of these data calculated using correlation, information gain, information gain ratio, and principal component analysis techniques. After these feature selection calculations, the results are evaluated by applying the methods of Support Vector Machines, Decision Trees, Naive Bayes, and Artificial Neural Networks from artificial learning techniques with the data sets created using both the original data set and the obtained features. The cross-validation method is used in all models and the results are evaluated with accuracy, classification error, AUC, recall, precision, and f-score performance indicators. According to the results obtained in this study, it is seen that artificial learning methods give successful results to diagnose types of anemia. Besides, feature selection methods increase the success in classification methods, but specific to this study, the artificial neural networks gave the most successful result with the original dataset.