Biologically Inspired Spiking CNN for Brain Tumor Classification

This research study focuses on the integration of Spiking Neural Networks (SNNs) and Convolutional Neural Networks (CNNs) in the development of a hybrid model known as Spiking Convolutional Neural Network (SCNN) for the purpose of brain tumor classification. Utilizing the distinct attributes of Spiking Neural Networks (SNNs), which replicate the spiking patterns observed in human brain neurons, along with the spatial feature extraction skills of Convolutional Neural Networks (CNNs), the proposed method intends to improve the precision and effectiveness of classifying brain tumors in medical images. This study begins with enabling a deep learning environment by importing necessary libraries, and loading the brain tumor dataset. This study employs the SCNN architecture, incorporating spiking convolutional layers for capturing complex patterns and details within medical images. The proposed model undergoes training with considerations for varying conditions, fostering adaptability. Visualization and analysis steps provide insights into the proposed model’s learning dynamics, activation distributions, and weight characteristics, ensuring a comprehensive understanding of its behavior. During the evaluation phase, the performance of the trained SCNN model is gauged by analyzing the accuracy and loss plots, showcasing its efficacy in detecting various tumor classes. Additionally, this study analyzes the temporal aspects introduced by SNNs, considering the timing of spiking events in neural activity. The proposed model’s efficiency is evaluated through training time per epoch analysis, considering the computational demands. The application of the proposed model for real-world scenarios is demonstrated through the predictions on new input data, showcasing its potential utility in clinical settings.