Enhanced Classification of Childhood Medulloblastoma Tumors Using ViT and Stain Normalization

This study presents a deep learning based approach for classifying Childhood Medulloblastoma tumors using medical imaging data, addressing the need for precise and timely diagnosis of the most common malignant brain tumor in children. Our research utilizes the pre-trained Vision Transformer (ViT) to demonstrate the impact of stain normalization on the medulloblastoma (MB) tumor Image classification dataset. Analyzing 151 histopathological images at 100x magnification, we show that using the Support Vector Machine (SVM) classifier after reducing the dimensionality of the trained features via Principal Component Analysis (PCA) enhances model performance compared to the standard fully connected (FC) layers classifier. The model achieves an accuracy of 94 %, precision of 97 %, and an F1 score of 93 %, highlighting its robustness. Our findings emphasize that combining Vision Transformer (ViT), transfer learning, and stain normalization significantly improves classification performance over traditional methods. This research underscores the potential of advanced deep learning techniques to enhance the diagnostic process for Childhood Medulloblastoma, offering a reliable and efficient tool for early and precise diagnosis.