Integrating multiclass classifiers for enhanced acute lymphoblast leukemia detection: A comparative study

Acute lymphoblastic leukemia (ALL) is a blood and bone marrow malignancy that is characterized by the growth of many immature lymphocytes known as lymphoblasts. It primarily affects children, particularly those aged two to five years, and is the primary cause of death in pediatric cancer cases. The method of treatment is determined to ALL, the individual?s age at the time of diagnosis, and other pertinent considerations. Regardless, early detection and diagnosis are critical for a good prognosis. It is critical to precisely detect malignant cells to make a diagnosis and assess the extent of the disease. However, due to physical similarities, identifying lymphoblasts from normal white blood cells under a microscope is often difficult. Using computer-aided techniques can be extremely valuable in automating the identification of cancerous cells, allowing histopathologists and oncologists to make decisions about the early stage. This paper demonstrates the usefulness of extensive image pre-processing, feature extraction from ResNet50 and VGG19 CNN models, and robust feature selection in an automated diagnostic technique for Acute Lymphoblastic Leukemia. Notably, on the C-NMC 2019 Dataset, ResNet50 with Random Forest feature selection appears as the best combination. The ResNet50 model achieves maximal precision, Weighted F1 Score, F1-score accuracy, and recall of 84.18%, 80.4%, 86.15%, 80.83%, and 88.7% respectively when combined with ANOVA and Random Forest. The combination of VGG19+Random Forest+SVM achieves a maximum accuracy of 86.2%. These findings highlight its exceptional performance in recognizing and categorizing target labels, demonstrating its ability to extract relevant properties for improved leukemia identification.