An Ensemble Based Convolutional Neural Network Modelling for Classifying Medulloblastoma Subtype

Medulloblastoma a heterogeneous pediatric brain tumor, requires specific subtype identification in order to determine the most effective treatment strategies. To evaluate the accuracy of classification, this study examines state-of-art leading Convolutional Neural Network (CNN), models like ResNet-50, MobileNet, Xception model, ensembles, etc in differentiating between medulloblastoma subtypes using histopathological images. The models undergo rigorous training and fine-tuning using a comprehensive benchmark dataset encompassing a variety of medulloblastoma subtypes. We also employ advanced image pre-processing techniques to optimize feature extraction. Through comparative analyses based on the accuracy, sensitivity, and specificity and precision of each model to uncover their unique strengths and weaknesses, the work identifies the top performing model for medulloblastoma subtypes classification. Then considering unique strengths of these models, we propose an ensemble methodology that combines top performing models for the underlying problem. Through experimentation, our ensemble approach proves to have the highest accuracy in medulloblastoma subtype classification. The combination of ResNet-50's depth, MobileNet's efficiency, and Xception's feature extraction abilities contribute to the ensemble's outstanding performance with 97% of acuuracy and F-score