ConvKAN: A Convolutional KAN for Brain Tumor MRI Classification

Brain tumors represent a critical challenge in medical diagnostics due to their diverse manifestations and the necessity for precise classification from MRI scans to facilitate effective treatment decisions. This paper introduces a novel Convolutional KAN (Kolmogorov Arnold Network) architecture tailored for the classification of brain tumors, including subtypes such as glioma, meningioma, pituitary adenoma, and non-tumor cases. Inspired by the Kolmogorov-Arnold superposition theorem, the network integrates advanced convolutional layers with attention mechanisms to enhance feature extraction and interpretability. The model harnesses the intricate spatial patterns and structural anomalies typical of brain tumors, capturing subtle nuances in MRI scans that traditional methods often overlook. Experimental validation demonstrates the network's robust performance in accurately categorizing tumors, leveraging saliency maps for visualizing crucial diagnostic features. This research significantly advances automated brain tumor diagnostics, potentially revolutionizing clinical practices by offering reliable and efficient diagnostic tools that aid in timely and accurate patient care.