The Brain Watching Itself: Identifying Brain Tumors with Spiking Neural Networks

This study investigates the application of Spiking Neural Networks (SNNs) in Magnetic Resonance Imaging (MRI) classification to explore their potential as an efficient, biologically inspired alternative to traditional convolutional models. Specifi-cally, we evaluate the performance of Leaky Integrate-and-Fire (LIF) neurons within a Convolutional Spiking Neural Network (CSNN). Additionally, we compare two encoding strategies- rate and temporal encoding- to assess their effectiveness in capturing MRI features. The temporal encoding CSNN achieved 73.10% accuracy, showing competitive performance with conventional CNNs while delivering up to 73% memory savings and 13 times faster inference. These results establish the viability of SNNs for MRI analysis, offering valuable insights into their computational efficiency and biological plausibility for resource-constrained clinical environments.