Enhancing ResNet Architectures with Bayesian Optimized Stochastic Configuration Networks for Lung X-Ray Image Classification

This paper introduces the ResNet-BOSCN architecture, which integrates Bayesian Optimized Stochastic Configuration Networks (SCNs) with traditional Residual Networks (ResNet) to tackle the challenges of lung X-ray image classification. Unlike conventional methods, ResNet-BOSCN employs Bayesian Optimization to dynamically optimize the network's hidden node count, effectively addressing the complexities inherent in medical imaging datasets. This optimization enables the architecture to adapt more precisely to varying data characteristics, significantly enhancing classification accuracy. ResNet-BOSCN achieves an accuracy of 98.81% in three-class lung X-ray image classification, marking a substantial improvement over the standard ResNet101’s accuracy of 97.62%.