<![CDATA[White blood cells are crucial components of the immune system responsible for combating infections and diseases. The classification and counting of white blood cells are typically performed manually by experienced operators or via automated cell analysis systems. The manual method is inefficient, time-consuming, and labor-intensive, while automated analysis machines are often expensive and require stringent sample preparation. This study aimed to compare the performance of three optimizers—root mean square propagation (RMSProp), stochastic gradient descent (SGD), and adaptive moment estimation (Adam)—in a white blood cell classification model using a convolutional neural network (CNN) algorithm. The dataset consisted of 12,392 images spanning four white blood cell classes: eosinophils, neutrophils, lymphocytes, and monocytes. The results indicate that the Adam optimizer achieved the best performance, with a training accuracy of 98.65% and an evaluation accuracy of 97.73%. Adam also outperformed the other optimizers in key metrics, including recall (97.43%), precision (97.42%), F1-score (97.42%), and specificity (99.11%). The AUC values for all classes exceeded 90%, demonstrating the model’s exceptional ability to distinguish between different cell types. The RMSProp optimizer yielded a training accuracy of 98.63%, whereas SGD achieved a lower training accuracy of 83.46%. This study highlights the significant impact of optimizer selection on CNN performance in white blood cell image classification, providing a foundational step toward the development of more accurate medical classification systems.]]>
Copyrights © 2025
