Next-Gen ST-CNN: A Lightweight Spatiotemporal Deep Learning Model for Accurate and Fast Brain Tumor Detection

Rapid and accurate brain tumor detection from MRI sequences is essential for timely diagnosis and intervention. This work presents a lightweight deep learning model that balances performance and computational efficiency. Methods: We propose Next-Gen ST-CNN, a spatio-temporal convolutional neural network that efficiently captures both intra-slice spatial features and inter-slice temporal dependencies. The model employs a novel decomposition of 3D convolutions to significantly reduce complexity while preserving diagnostic features. Training and evaluation were performed on a standard brain tumor MRI benchmark dataset, covering glioma, meningioma, pituitary, and non-tumor cases. Results: The proposed model achieves a test accuracy of 98.63%, along with macro precision, recall, and F1 scores ranging from 97% to 100%. It records a low test loss of 0.069 and a fast training time of 367.2 seconds. With only 2.17 million parameters, the model is compact yet powerful. Conclusion: Next-Gen ST-CNN effectively integrates spatial and temporal analysis while maintaining a lightweight architecture, making it suitable for real-time deployment. Significance: The proposed approach offers a compelling solution for next-generation computer-aided diagnosis in neurooncology, enabling fast, reliable and accessible brain tumor screening using minimal resources.