OkanNet: A Lightweight Deep Learning Architecture for Classification of Brain Tumor from MRI Images
New custom CNN architecture achieves 88.1% accuracy on 7,023 MRI images while being optimized for mobile deployment.
Researchers Okan Uçar and Murat Kurt have introduced OkanNet, a novel lightweight Convolutional Neural Network (CNN) designed specifically for classifying brain tumors from MRI scans. The study, detailed in a new arXiv preprint, presents a comparative analysis between this custom-built architecture and a Transfer Learning approach using the established ResNet-50 model. The goal was to balance high diagnostic accuracy with computational efficiency, a critical consideration for real-world clinical deployment.
In experiments conducted on an extended dataset of 7,023 MRI images, the ResNet-50 model demonstrated superior classification performance, achieving 96.49% accuracy and 0.963 precision in distinguishing between Glioma, Meningioma, Pituitary tumors, and no tumor. However, the custom OkanNet architecture, while reaching a lower accuracy of 88.10%, proved to be approximately 3.2 times faster during training, completing the process in just 311 seconds. This significant speed advantage highlights its design for low computational cost.
The research underscores a fundamental engineering trade-off in medical AI: depth and complexity for peak accuracy versus lightweight design for speed and accessibility. OkanNet is positioned not as a replacement for high-performance models in well-resourced settings, but as a strong alternative for mobile, edge, and embedded systems. These could include portable diagnostic tools or applications in regions with limited computing infrastructure, where rapid, on-device analysis is paramount.
- OkanNet, a custom lightweight CNN, achieved 88.1% accuracy classifying brain tumors from MRI scans.
- It trained 3.2x faster than ResNet-50 (311 seconds vs. ~995 seconds), trading ~8% accuracy for speed.
- Designed for mobile/embedded systems, it demonstrates a practical speed-accuracy trade-off for deployable medical AI.
Why It Matters
Enables faster, on-device AI diagnostics for brain tumors in resource-constrained clinical and mobile settings.