A Deep Learning Approach for Diagnosis and Detection of Acute lymphocytic leukemia

ALL is a fast-growing condition that can impact the bone marrow, blood, and crucial organs within the body. It is a rapidly progressive disorder that requires prompt medical attention and treatment. Timely and accurate diagnosis is crucial to determine appropriate treatment. However, the manual diagnosis by hematologists can be time-consuming and susceptible to errors and inconsistencies. Deep learning, particularly convolutional neural network (CNN/ConvNet) technology, is used to address various challenges in detecting and diagnosing ALL. This advanced technology is capable of accurately analyzing large amounts of data and identifying subtle differences in the genetic and cellular makeup of leukemia cells, has shown promise in medical diagnosis, particularly in visual image analysis. The fast and accurate feature extraction function and trainable network architecture of CNNs make them suitable for this application. In this study, we propose an effective deep framework for CNN networks that enables the rapid and early detection and classification of leukemia cells from microscopic blood smear images. In order to classify cells as either blast cells (ALL) or healthy cells (HEM), the problem was framed as a binary classification challenge. Each individual cell was assessed and classified using this approach. In this study, various CNN-based architectures were evaluated using transfer learning. Specifically, we investigated the performance of VGG16, VGG19, ResNet60, ResNet101, and DensNet169 models. The results indicate that the DensNet169 model outperformed the other models with an accuracy score of 97.62% . These findings demonstrate the potential of transfer learning and the superiority of the DensNet169 model for the accurate and efficient classification of leukemia cells in microscopic blood smear images.