Revolutionizing Leukemia Diagnosis: Fine-Tuned VGG19 CNN for Enhanced Classification Accuracy

The complicated haematological cancer leukemia calls for precise and quick diagnosis techniques to increase patient outcomes. The VGG19 Convolutional Neural Network (CNN) architecture is explored in this work for leukemia classification from high-resolution blood pictures. With a dataset of 7,500 photos split into four categories—"Benign," "Malignant Pre-B," "Malignant-B," and "Malignant early Pre-B"—the study seeks to improve automated leukemia detection by use of modern image processing and deep learning methods. Renowned for strong feature extraction, the VGG19 model is optimized with bespoke classification layers targeted for leukemia diagnosis. Early stopping and learning rate scheduling help to maximize performance by means of the Adam optimizer and categorical cross-entropy loss function in the training process. With accuracy, precision, recall, F1-score, and confusion matrix among other evaluation measures, the tuned VGG19 model shows an overall accuracy of 99%. With modest classification mistakes in "Benign" and "Malignant early Pre-B" instances, the model exhibits extraordinary performance in separating between leukemia subtypes. This work validates that the VGG19-based CNN method greatly improves automated leukemia classification, so offering a trustworthy instrument for raising diagnosis accuracy and helping medical practitioners in leukemia treatment.