Machine Learning Approaches for the Detection and Classification of Leukemia from Histopathological Images

Leukemia is a severe and potentially life-threatening ailment affecting the blood and bone marrow. Timely detection is of utmost importance as it can greatly enhance patient prognosis. In recent years, deep learning techniques have shown remarkable promise in the realm of medical image analysis. The primary aim of this project is to establish a deep learning framework tailored for the early identification of leukemia cancer cells within microscopic blood samples. Image dataset of blood samples were collected and pre-processing was done. For feature extraction VGG19 was used. The features were employed for both training and testing purposes across a range of machine learning classification algorithms, including Random Forest, KNN, NBC, Decision Tree, and SVM-RBF. The most effective one can be determined by assessing the classifiers' performance through metrics like accuracy, precision, etc. Here in the proposed model where we have two approaches – one with VGG19 used for feature extraction and the other without using VGG19. Without the VGG19, Random Forest classifier had the highest accuracy – 77.9%. With VGG19 used for feature extraction, SVM-RBF classifier performed well with an accuracy of 85.1%.