Medical Image Classification with Spiking Neural Networks

Our research explores energy-efficient spiking neural networks for privacy-preserving medical applications. Inspired by biological neural mechanisms, these networks offer significantly lower power consumption compared to conventional artificial neural networks, making them ideal for resource-constrained applications. Their compact architectures reduce memory usage and computational demands while maintaining strong performance on complex AI tasks.

The research activity addresses critical healthcare challenges, considering energy consumption and privacy issues. By enabling distributed model training without centralizing sensitive patient data, the approach preserves privacy while leveraging diverse datasets. low energy consumption makes this kind of networks suitable for wearable devices.

The research methodology includes: analysing existing spiking network models and their training mechanisms; implementing and evaluating models on medical datasets; developing optimized variants through ablation studies tailored to medical tasks; exploring various training strategies and comparing performance with traditional approaches; and validating generalizability across larger, more diverse datasets.