Erythrocyte Classification Using Alexnet and Simple Cnn

One examination method to support thalassemia diagnosis is a blood morphological examination of the patient's peripheral blood smear. However, manual analysis of peripheral blood smears requires time, unique expertise, and expert eye fatigue. This paper proposes a computer vision method using deep learning to assist experts in examining peripheral blood smears. The dataset consists of nine erythrocyte types that appear in thalassemia patients. The image size normalization was conducted before the deep learning model used the image. Data augmentation was used to increase the number of data in the datasets. The transfer learning approach is used to improve classification results. The erythrocyte classification result using Alexnet and simple CNN has been compared. The best performance of the Alexnet model reached 95.92% accuracy, 91.46% sensitivity, and 99.48% specificity.