AI-Supported Pneumonia Diagnosis: Performance of Deep Learning Models and Synthetic Data Generation

Pneumonia, a respiratory infection that affects the lungs and leads to serious health complications, is diagnosed by physicians based on symptoms, physical exam findings and imaging tests such as chest X-rays, but these methods can be time-consuming and subjective. In this study, we propose a Convolutional Neural Network (CNN) based pneumonia disease detection model utilizing X-ray images for accelerating diagnosis and initiate treatment early to prevent disease progression. Due to the insufficient number of original X-ray samples, we also produced synthetic images with our developed Deep Convolutional Generative Adversarial Network (DC-GAN) model. We comparatively evaluated the performance of our models with pretrained VGG16 and ResNet50 models on synthetic and original image datasets separately and combined. Our results show that all models performed best with the synthetic dataset and also the combined set expanded with synthetic data gave better results in all cases than using the original data alone. The average test accuracy results of all models were obtained as $93 \%$, $89 \%$ and $91 \%$ with synthetic, original and combined datasets, respectively. Thus, it proves that the good quality of the produced synthetic images that can be used to solve the problem of medical image data need, especially in artificial intelligence-based decision support systems. When the models were compared, our CNN model achieved the best performance in all scenarios with the accuracy values of $95 \%, 91 \%$ and $93 \%$ with synthetic, original and combined datasets, respectively.