Epilepsy is a neurological condition marked by epileptic seizures. Focal cortical dysplasia (FCD) is a leading cause of drug-resistant epilepsy. Surgical removal of FCD lesions is the most effective treatment, which is heavily dependent on their precise localization and delineation. 

However, identifying FCD lesions in magnetic resonance (MR) images remains challenging in clinical practice due to the subtle structural changes they cause.

In a study published in Insights into Imaging, a team led by Dr. XU Jinping from the Shenzhen Institute of Advanced Technology of the Chinese Academy of Sciences, together with collaborators, proposed a multiscale transformer-based model for the end-to-end segmentation of FCD lesion from multi-channel MR images, enhancing the feature representation of lesions in global field of view.

The model integrates a convolutional neural network (CNN)-based encoder-decoder structure with multiscale transformer pathways. The CNN encoder extracts local features, which are then fed into respective transformer pathways to capture global features at various scales. 

To reduce complexity and prevent overfitting, researchers utilized a computation- and memory-efficient Dual-Self-Attention (DSA) module to construct the transformer pathway. The DSA module consists of a spatial branch and a channel branch, which identified long-range dependencies between feature positions and channels, thus highlighting areas and channels pertinent to lesions.

Besides, researchers trained and evaluated the proposed model on a public dataset of MR images from 85 patients, using both subject-level and voxel-level metrics.

Experimental results showed that the method successfully identified lesions in 82.4% of patients, with a low false-positive lesion cluster rate of 0.176±0.381 per patient. Also, the model achieved an average Dice Coefficient of 0.410±0.288, surpassing five established methods.

"As far as we know, this is the first study to apply a transformer-based model for the FCD lesion segmentation," said Dr. XU. "Our study promises to be a valuable tool for medical practitioners, enabling them to detect FCD lesions swiftly and accurately."