Comparative Study of Different CNN Architectures for B-ALL Diagnosis from Peripheral Blood Smear Images
Unneta Chatterjee, Hreetam Paul, Kanik Palodhi · 2024
Accurate diagnosis of acute lymphoblastic leukemia (ALL) often relies on invasive and costly procedures. This paper explores the use of convolutional neural networks (CNNs) to classify ALL subtypes from peripheral blood smear (PBS) images, offering a less invasive alternative. We evaluated ResNet152, DenseNet201, and a sequential DenseNet-ResNet model on a dataset of $\mathbf{3, 2 5 6}$ PBS images from 89 patients. The sequential model demonstrated superior performance in accuracy and robustness. Our findings suggest that advanced CNN models can enhance ALL diagnosis and subtype classification, potentially reducing diagnostic errors and improving possibility of early detection.