Harnessing generative AI for enhanced brain tumor detection in clinical trials

Abstract

Artificial intelligence (AI) in integrating its method of generative models in the healthcare sector has proven beneficial in enhancing diagnostic efficiency and, subsequently, patient experience. The major limitation relevant to the proposed approaches for detecting brain tumor problems is that various and sufficient datasets for brain tumor detection are rare. This research seeks to overcome this parameter by applying and enhancing the application of generative AI in medical imaging, particularly generative adversarial networks (GANs) in data augmentation in the diagnosis of brain tumors. It also uses GANs for data augmentation, specifically generating synthetic magnetic resonance imaging (MRI) images to enhance brain tumor detection. Our study introduces a hybrid model that combines GANs for realistic image generation with a pre-trained ResNet50 architecture for feature extraction and classification. By employing GANs, new, diverse MRI samples are generated, significantly expanding the available dataset. The augmented dataset is used to train the hybrid model, resulting in high accuracy and AUC-ROC score, demonstrating a substantial improvement over the traditional method. Key metrics such as precision and recall also indicate the model's enhanced ability to distinguish between tumor and non-tumor images. This approach highlights the potential of GAN-based data augmentation to improve generalization in medical imaging, addressing the scarcity of large, varied datasets. The findings contribute to advancing AI-driven diagnostic tools in healthcare, with the potential to enhance early diagnosis and treatment outcomes for brain tumor patients. However, ethical concerns related to data privacy, bias, and the responsible deployment of AI in clinical practice remain important challenges that must be addressed in future work.

Ähnliche Arbeiten