Convolutional Neural Network-Based Approach for Classification of Blood Cells in Leukaemia Detection

Blood cell malignancies come in a variety of forms, including lymphoma and leukaemia, and because of their varied cellular and molecular presentations, they are challenging to diagnose and treat. The majority of conventional diagnostic methods rely on cytological testing and genetic profiling, which can be erratic and time-consuming. Recent developments in deep learning and computer vision offer great potential to automate the processing of medical imaging data, hence improving the speed and Accuracy of diagnosis. Current work investigates the application of convolutional neural networks for blood cell cancer subtype classification based on microscopic image analysis. The study makes use of a sizable dataset that has images categorized into four classes: benign, malignant, and its subtypes, malignant Pre-B, malignant Pro-B, and malignant Early Pre. The dataset is pre-processed and split into training and validation subsets to improve model generalization and test sets, and then augmentation techniques are used. To extract valuable characteristics from the images, a custom CNN architecture, including convolutional layers, batch normalization, max pooling, and dropout layers, is created and trained. Stochastic gradient descent (SGD) is used to optimize the model, and measures like Accuracy, Precision, Recall, and F1-score are used to assess the model. Confusion matrices and accuracy and loss curves help visualizers clarify the learning mechanisms and performance of the model. The result shows that deep learning could advance oncology diagnostic capabilities by being able to distinguish between several blood cell cancer subtypes with promising performance. The results add to the continuing attempts to use artificial intelligence for haematologic malignancy early identification and personalized treatment plans.