Detection and Classification of Acute Lymphoblastic Leukemia Utilizing Deep Transfer Learning

Amutation in the DNA of a single cell that compromises its function initiates leukemia. This leads to the overproduction of immature white blood cells, which encroach upon the space required for the generation of healthy blood cells. Leukemia is treatable if identified in its initial stages. Nonetheless, its diagnosis is both arduous and time-consuming. In this study, a novel approach for diagnosing leukemia across four stages—Benign, Early, Pre, and Pro—utilizing deep learning techniques. Firstly, we had employed two Convolutional Neural Network (CNN) models: MobileNetV2 with an altered head and a bespoke model. The custom model has multiple convolutional layers, each paired with corresponding max pooling layers. Nowwehave employed two more pretrained model: InceptionV2 and VGG16 with altered heads. We utilized MobileNetV2, InceptionV2 and VGG16 with ImageNet weights, and the head was adjusted to integrate the final results. Finally, we have utilized ensemble with soft voting technique to comprehend the results from multiple neural networks. The utilized dataset is a publicly available collection of blood cell smear images titled “Acute Lymphoblastic Leukemia (ALL) image dataset”, and then used the Synthetic Minority Oversampling Technique (SMOTE) to augment and balance the training dataset. Which attained an accuracy of 96.34% with the custom model, while MobileNetV2 and InceptionV2 achieved a superior accuracy of 99.39%. The VGG16 achieved an accuracy of 99.08%. Finally, the ensemble technique ensured more promising result with 99.70% accuracy. The pre-trained model exhibited encouraging results and an increased likelihood of real-world application.