Region-Guided Generative Adversarial Networks for Small Sample MRI Generation from CT in Temporal Lobe Epilepsy

As one of the most common neurological disorders, epilepsy poses major challenges in clinical practice. Temporal lobe epilepsy (TLE), which is highly prevalent among drug-resistant cases, often fails to respond to anti-epileptic drugs (AEDs), making surgical intervention a necessary treatment option. Consequently, accurate identification of the epileptogenic zone is essential to optimize surgical outcomes. MRI is the most widely used non-invasive imaging technique for epileptogenic focus localization, offering superior soft-tissue contrast and diagnostic accuracy. Nevertheless, the utility of MRI can be restricted due to time limitations, high costs, or certain patient-specific contraindications. Recently, the deep learning breakthroughs have made the synthesis of MRI images from CT scans possible. However, the patients undergoing both CT and MRI scans are limited, making it difficult to obtain large-scale paired datasets. In this study, we have proposed an RSPGAN method which is based on generative adversarial networks (GANs). This method has integrated image registration and temporal lobe segmentation. In the result, the images generated by the RSPGAN method achieve an SSIM of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.97 \pm 0.02$</tex> and a PSNR of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$30.36 \pm 4.74$</tex>, with significantly reduced error metrics: MAE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$4.93 \pm 2.82$</tex> and MSE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$79.10 \pm 43.49$</tex>, outperforming the other models. The experimental results demonstrate that our RSPGAN method achieves superior performance on limited datasets.