Abstract
Despite extensive research on conventional deep learning architectures, there is a notable gap in the exploration of Convolutional Kolmogorov-Arnold Networks for real-world image classification tasks. This study investigates the application of a custom Convolutional KAN architecture for lung cancer subtype classification using histopathological images—a critical step for early diagnosis and effective treatment planning. The approach described in this research paper achieved 96% accuracy, supported by optimizations such as vectorization during training. For comparative analysis, baseline models including a custom CNN and a ResNet-18 were also evaluated. The promising performance of the Convolutional KAN model underscores the potential of less-explored architectures in advancing medical image classification.