Performance Comparison of SqueezeNet Implementing a Dendritic Neural Model for Brain Disease Image Classification

In this study, a comparative analysis is conducted between a pre-trained SqueezeNet v1.1 model and a modified SqueezeNet architecture integrating a dendritic neural model (DNM) into the classifier layer for eight-class brain disease classification using MRI images. The proposed modification replaces the standard linear classifier with a dendritic-based processing mechanism to enhance nonlinear representation at the classification stage. Experiments are performed on an MRI-based brain medical image dataset, and model performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. The results show that the SqueezeNet model integrated with DNM achieves a significant accuracy improvement of approximately 5%, increasing from 90.33% to 95.61%, compared to the standard SqueezeNet model. This performance gain is accompanied by a moderate increase in model complexity, with the parameter count rising by approximately 1.69% (from 726.6K to 738.9K) and a longer convergence time (40 epochs versus 22 epochs). Overall, the findings indicate that incorporating DNM into a lightweight CNN architecture such as SqueezeNet can effectively improve medical image classification performance while maintaining reasonable computational efficiency.