Deep learning based Synthetic Magnetic Resonance Image Generation from Computed Tomography Imaging for Cerebral Ischemic Stroke Patient

Cerebral ischemic stroke is a leading cause of death and disability worldwide, and timely diagnosis and treatment are crucial for improving patient outcomes. Computed tomography (CT) and magnetic resonance (MR) imaging are both widely used imaging modalities for diagnosing stroke. However, CT is often the first choice due to its accessibility and speed, while MR provides higher sensitivity and specificity. This study aims to investigate the feasibility of converting CT images of cerebral ischemic stroke patients to MR images using deep learning techniques. The proposed method utilizes a Generative Adversarial Network(GAN) to learn the mapping between the two modalities and generate synthetic MR images from CT scans. We have added a new attribute to the network, based on the observation that stroke lesions are highly discernible on MR images. The performance of the method is evaluated on a dataset of stroke patient images. Experimental results demonstrate that the proposed approach can effectively generate high-quality MR images from CT scans, achieving a peak signal-to-noise ratio (PSNR) of 24.4 dB and a structural similarity index (SSIM) of 0.78. Furthermore, our method outperforms existing image-to-image translation methods in terms of visual quality and quantitative evaluation metrics.