Fourier Domain Adaptation for Image Augmentation in CNN-based Pneumonia Classification

Pneumonia remains a significant global health concern, necessitating accurate and efficient diagnostic tools. Convolutional Neural Networks (CNNs) have demonstrated promise in medical image classification, including pneumonia detection, but are often limited by small datasets and variations in imaging conditions. This paper introduces a novel approach for pneumonia image classification that leverages Fourier Domain Adaptation (FDA) to augment the dataset and enhance CNN performance. This proposed method combines the power of CNNs with Fourier domain transformation, effectively adapting the network to pneumonia-specific features. This adaptation enriches the dataset by generating diverse representations of pneumonia patterns, addressing the challenges of data scarcity and model generalization. Experimental results showcase the effectiveness of this approach, with training accuracy reaching 93.75%, validation accuracy achieving a perfect score of 100%, and robust testing accuracy of 98.98%. These remarkable results underscore the significant impact of Fourier domain adaptation in improving pneumonia classification accuracy and reducing misclassification errors. By combining the capabilities of CNNs with Fourier domain adaptation, this proposed method offers a promising avenue for enhancing pneumonia diagnosis, potentially leading to more accurate and timely medical interventions. This research contributes valuable insights to the field of medical image classification and holds great potential for improving patient care and health outcomes.