Comparative Evaluation for Classification of WBC Sub-Particles Using Google Colab GPU/TPUs

This study focuses on improving manual classification of medical laboratories for WBC Sub-particles such as Eosinophil, Lymphocyte, Monocyte, Neutrophil. This study compares A100, TPUv2, and T4 GPU/TPUs against its performance in classifying WBC using Convolutional Neural Network (CNN) and Support Vector Machine (SVM). Datasets from opensource were validated for veracity by a medical diagnostic center before training, testing, and evaluation of the model. Manual verification of the classification results was conducted with the same agency. There are 4000 datasets for WBC classification (1000 each, 70% training set, 30% test set) using CNN, and at least 330 datasets (70% training set, 30% training set) using SVM. Results showed that the combination of CNN-A100 and SVM-A100, with 97.13% and 87.16% respectively, yielded the highest accuracy and lowest loss, showing that A100 has the most balanced architecture and performance metrics for this study.