Convolutional Kolmogorov-Arnold Network for Pneumonia Detection in Medical Image Analysis

Pneumonia is a serious respiratory infection that poses a significant global health burden, particularly in regions with limited access to medical personnel and diagnostic resources. Chest X-ray imaging remains the most common method for pneumonia diagnosis, however, manual interpretation is prone to error and often requires experienced radiologists. To address this challenge, automated diagnostic systems based on deep learning have gained increasing attention. This study aims to evaluate the effectiveness of the Convolutional Kolmogorov-Arnold Network (CKAN) in detecting pneumonia from chest X-ray images and compare its performance against a baseline Convolutional Neural Network (CNN) model. The study involved three variations of CKAN architecture that combined convolutional layers with Kolmogorov-Arnold-based layers. Both CKAN and CNN models were trained on balanced and imbalanced datasets using data augmentation techniques to improve model robustness. Additional experiments were conducted with and without the application of early stopping mechanisms. Performance evaluation was conducted using five metrics: accuracy, precision, recall, specificity, and balanced accuracy. Loss history and confusion matrices were also analyzed to assess learning stability and classification behavior. The best-performing CKAN model achieved an accuracy of 83.49%, precision of 79.96%, recall of 98.21%, specificity of 78.59%, and balanced accuracy of 78.59%. In comparison, the best-performing CNN model reached 81%, 77.98%, 97.18%, 75.73%, and 75.73%, respectively. These results demonstrate CKAN’s superior generalization capability and its effectiveness in handling class imbalance. In conclusion, CKAN shows promising potential for improving pneumonia detection from chest X-rays using a more compact and interpretable model structure. Future studies can explore hyperparameter optimization and extend the method to other medical imaging tasks. This work contributes to the development of more accurate and accessible automated diagnostic systems.