Binary and Multiclass Leukemia Classification Using a Hybrid Deep Learning-ML Approach: Towards an Explainable AI Framework

Within the latter phases of the disease, the spreading of leukaemia in people generates a great deal of concern. Blood production in the bone marrow is hindered as a result of this. For the purpose of diagnosis in this instance, morphology analysis of blood cells from people is a method that is well-known and has been well tested. Both normal cells and cells that have been infected with leukaemia may be differentiated using the binary categorization. In addition, the therapy for various types of leukaemia is distinct from one another. For a correct diagnosis of the kind of leukaemia, it is also necessary to identify these subclasses. The determination of the leukaemia subtype requires the use of multiple classifications. Microscopical inspection of these blood cells is often used to accomplish this task. A further reason why the procedure for making choices is so important is that it necessitates the presence of a skilled pathology. This led to the creation of a software-based structure for diagnosis. The investigators used methods of machine learning that are considered to be state-of-the-art. These methods include support vector machine (SVM), random forest (RF), naïve bays, k-nearest neighbour (KNN), and others. However, these methods result in the accuracy of classification that is restricted. Researchers also make use of more sophisticated varieties of deep learning techniques. The suggested XGBoost approach outperformed all models with 97% accuracy. DenseNet-XGBoost and Xception-RF performed well, but Inception-ResNet and Random Forest(RF) performed moderately. ResNet-RF was lowest. Thus, the suggested leukemia categorization method works best.