White Blood Cell Classification using Vision Transformer Model

Deep learning techniques present significant potential for automatic classification of white blood cells (WBC) in blood smear images, which are very well assisted to get better medical diagnostic solutions. However, the distribution of different types of WBCs is often imbalanced, leading to bias in model training and difficulties in generalizing, and along with this, some WBC types have similar morphological features, making it challenging for the model to accurately distinguish between them. To address these challenges, we proposed a ViTB/32-based transformer model because transformer models are resource-intensive and works parallelly to execute inputs. therefore, the diagnostic model's accuracy of clinical features enhances effectively and improves the computational complexity. This study compares our proposed model with other leading SOTA transfer learning models like ResNet50V2, EfficientNetB7, DenseNet201, and InceptionV3, comparing based on performance metrics, revealing their strengths and limitations for WBC classification. The Vision Transformer model (ViT-B/32) outperforms other transfer learning-based models in terms of accuracy (96.06%), recall (93.2%), precision (89.6%), and F1-score (91.5%), indicating its superior ability to classify WBC images correctly.