Automatic detection of FCDs with AI
Automated detection and segmentation of focal cortical dysplasias (FCDs)
Together with experts from the Epilepsy Center Hamburg, we have written a Research Paper about the automated detection of FCDs with Artificial Intelligence. In the article called “Automated detection and segmentation of focal cortical dysplasias (FCDs) with artificial intelligence: presentation of a novel convolutional neural network and its prospective clinical validation.” we present the results of our 2 years of research about applying AI to help to solve the challenges in the field of epilepsy. In this post, we would like to share with you the basic information about it.
Our algorithm can contribute to better detection of FCDs.
Focal cortical dysplasias (FCDs) represent one of the most frequent causes of pharmaco-resistant focal epilepsies. Despite improved clinical imaging methods over the past years, FCD detection remains challenging, as FCDs vary in location, size, and shape and commonly blend into surrounding tissues without clear definable boundaries. This fact causes that only experts with many years of experience are able to reliably detect the FCDs. To face this problem, we developed a novel AI-based solution to automatically detect and segment FCDs on MRI pictures to contribute to better detection of them.
Work with field experts
Training AI models in such complicated fields as detection of FCDs, where the malformations of the brain are very hard to find, require tight cooperation with the field experts. We are working on this with a group of epileptologists, lead by Dr. Patrick House from the Epilepsy Center in Hamburg, who has over 20 years of experience specifically working with FCDs. The cooperation with the doctors allowed us to collect the largest FCD training dataset to date with various types of FCDs and some focal PMG, as well as run many detailed validation rounds to verify the results of the algorithm.
How was the artificial intelligence trained?
The neural network has been trained on 201 T1 and FLAIR 3 T MRI volume sequences of 158 patients with mainly FCDs, regardless of type, and 7 focal PMG. For the purpose of training, we have used also 100 normal MRIs and 50 MRIs with other pathologies than FCD/PMG. The doctors we work with applied the algorithm prospectively on 100 consecutive MRIs of patients with focal epilepsy from daily clinical practice. The results were compared with corresponding neuroradiological reports and morphometric MRI analyses evaluated by an experienced epileptologist.
The algorithm is already considered useful in clinical practice for FCD screening
The architecture of the solution was based on 3D convolutional neural network with autoencoder regularization for FCD detection and segmentation. Our algorithm provided a higher sensitivity in detecting FCDs than conventional visual analyses. Despite its low specificity, the number of false positively predicted lesions per MRI was lower than with morphometric analysis. Best training results reached a sensitivity (recall) of 70.1 % and a precision of 54.3 % for detecting FCDs. Applied on the daily-routine MRIs, 7 out of 9 FCDs were detected and segmented correctly with a sensitivity of 77.8 % and a specificity of 5.5 %. The results of conventional visual analyses were 33.3 % and 94.5 %, respectively (3/9 FCDs detected); the results of morphometric analyses with overall epileptologic evaluation were both 100 % (9/9 FCDs detected) and thus served as reference.
The epileptologists consider our algorithm already useful for FCD pre-screening in everyday clinical practice, whereas the solution is constantly being improved.
“Many FCDs as a source of patients’ epilepsy often go undetected. It is important that focal cortical dysplasias are detected because they can be operated on. ” said Dr. Patrick House “The algorithm is therefore very promising. Furthermore, in addition to epileptologists, it can help radiologists or neurologists detect FCDs on MRIs.”