Enhancing diagnostic precision: machine learning for blood cell image classification

Abstract

The accurate and timely analysis of blood cell images plays a crucial role in disease diagnosis and monitoring. This research looks into the segmentation and classification of white blood cells using machine learning techniques. The process entails meticulously collecting a dataset of 12,500 photos covering four critical cell types. To enhance critical information, the picture preprocessing stage employs complicated operations such as color space conversion, blurring, threshold setting, contour detection, and overlay techniques. A variety of models are investigated, including EfficientNetB3, Vgg16, VGG19, Inception v3, and MobileNet v2, with VGG16 appearing as the best choice. An ablation study is used to further investigate the impact of activation functions, hidden units, learning rates, and batch sizes on the model's performance. The final model configuration is 96.96% accurate, with detailed statistical indicators enabling a nuanced assessment of its strengths and limits. The ablation investigation reveals the model's susceptibility to certain configurations, guiding the selection of ideal parameters. This work advances blood cell image categorization by providing insights into model behavior and setting the path for future improvements, such as dataset expansion and potential incorporation into clinical workflows.