Intelligent Multi-Model Deep Learning Approach for Accurate Leukaemia Diagnosis and Stage Classification

Abstract

Leukaemia, a malignant disorder of the blood and bone marrow, requires early and accurate detection to improve patient survival and treatment planning. Traditional manual examination of peripheral blood smears is labor-intensive, prone to errors, and subject to observer variability. This study proposes an automated framework for leukemia detection and stage prediction using multi model convolutional neural networks with transfer learning. Four pre-trained architectures—ResNet50, DenseNet121, MobileNetV2, and EfficientNetB3are fine-tuned on microscopic blood smear images to classify healthy and leukemic cells as well as predict disease stages. Each model is evaluated across standard performance metrics, with accuracy considered the primary benchmark for model selection. Experimental results reveal that among the tested architectures, this model achieving the highest accuracy emerges as the optimal choice for reliable leukaemia detection and stage prediction.