Hamburg Epilepsy Center has the largest monitoring station in Germany for comprehensive epilepsy diagnostics, thus being one of the most important centers for detecting and curing epilepsy in Germany and in Europe.
As FCDs may vary in location, size, and shape and commonly blend intro surrounding tissues without clear definable boundaries, its detection remains a very challenging task also for medical professionals and reliable detection of FCD requires evaluation by an experienced epileptologist. This fact causes that many FCDs as the source of epilepsy remain undetected, whereas detected FCDs can be operated out and the patient can be free of epilepsy seizures.
Together with the experts from Epilepsy Hamburg, we decided to use the power of artificial intelligence to automatically detect and recognize the position of FCD on medical images. We developed a 3D convolutional neural network with autoencoder regularization for FCD detection and segmentation. It provided higher sensitivity in detecting FCDs than conventional visual analyses. The developed model is considered useful by the experts for FCD screening in clinical practice.
Recognition of FCDs is a much more complicated task than many other appliances of AI in radiology, such as segmenting brain tissues or detection tumors. The substantial difficulty is caused by the fact that FCDs vary greatly in location, size, and shape and mostly blend into surrounding tissue without forming clear definable boundaries. The MRI analysis done by the experts is time-consuming, requires expertise and in some cases remains subjective which makes it harder to collect a reliable dataset for the training of AI. Thanks to extensive technological research and the great work of the doctors from the Epilepsy Center we managed to create the biggest dataset of MRI with FCDs collected so far and achieve state-of-the-art results in FCD detection.