Enhanced chronic kidney disease detection using auxiliary classifier GANs and hybrid data augmentation on CT imaging

Chronic kidney disease, which can be caused by kidney cysts, tumours, and stones, is difficult to detect in its early stages due to the absence of clear symptoms. Deep learning models show promise in predicting this condition using kidney computed tomography scans. However, their effectiveness in real-world applications is often limited by the imbalance in publicly available medical image datasets. Our research uses a picture archiving and communication system with computed tomography to address this limitation. The 12,446 images in the kidney normal, cysts, tumour, and stone dataset from Kaggle were divided across four categories. To overcome disparities in class, we explored multiple data augmentation methods, such as synthetic minority oversampling, and generative adversarial networks. Our approach improved the model’s generalisation by ensuring an equal and balanced representation of varied kidney states through these methods. On the original dataset, VGG16 achieved 99.2%, EfficientNetV2 achieved 97.62%, and MobileNetV2 achieved 97.2%. On the ACGAN-augmented dataset, these adjusted to 97.3%, 97.0%, and 95.1% respectively. The results shown here make it easy to comprehend how the use of artificial data creation techniques enhances the performance of deep learning models in the context of medical applications. This study indicates the potential of synthetic data production methods to improve the performance of deep learning models in clinical diagnostic contexts. It also highlights the necessity of employing well-balanced datasets, which is an important aspect of the field of medical image analysis.