A robust hybrid deep learning model for acute lymphoblastic leukemia diagnosis
Bakkanarappa Gari Mounika, Mohammad Faiz, Nausheen Fatima, Ramandeep Kaur Sandhu · 2024
Acute lymphoblastic leukemia (ALL), a rare kind of blood cancer, can strike young children and adults. It has a strong possibility of being cured if caught early. The study’s objectives are to decrease the time needed for identification, increase diagnosis accuracy, and lower the cost of specialized care for ALL. The ALL-IDB1 data, which includes 108 images of lymphocyte cells, was used in the study. By using image augmentation to apply transformation parameters to source images, the Keras library created 3240 new images. After extracting features from augmented data using the MobileNetV2 model, the XGBoost classifier was trained to predict the labels that would go with each feature. For the ALL-IDB1 dataset, the proposed approach, MobileNet V2+XGBoost, is compared with other cutting-edge models such as GoogLeNet, ResNet50, & MobileNet V2+SVM. The proposed method had a 99.07% accuracy rate, a 99.35% precision rate, and a 98.72% recall rate. MobileNet V2+XGBoost has successfully trained to generalize and outperform on the substantial amount of augment data. These findings showed the possibility of the proposed approach for giving a reliable ALL diagnosis, which could ultimately result in better patient outcomes.