Project Summary:
Epilepsy remains a significant public health challenge, notably drug-resistant focal epilepsy, which affects approximately 30-40% of those diagnosed. Despite advances in pharmacological treatments, a substantial portion of these patients fail to achieve seizure control with medications alone, necessitating invasive surgeries. Unfortunately, these surgeries still yield high recurrence rates—approximately 50% in temporal lobe cases and 70-80% in frontal lobe cases. These statistics highlight a critical need for more precise identification and excision of epileptogenic zones to enhance surgical outcomes and improve patient quality of life.
In Wisconsin, the rise in drug-resistant epilepsy cases over the past decade mirrors a broader national trend that significantly burdens healthcare systems. This increasing prevalence underscores the necessity for innovations that enhance surgical precision and efficiency, thereby aligning with state and national health priorities to reduce long-term healthcare costs and improve patient outcomes.
Magnetoencephalography (MEG), a non-invasive technique that measures magnetic fields generated by neuronal activity, has shown great promise in precisely mapping epileptogenic zones, correlating well with intracranial EEG (iEEG) findings. Traditionally, the analysis of MEG for identifying epileptic discharges has depended on clinician interpretation, a process that is not only time-consuming but also fraught with potential biases. Leveraging advanced, automated deep learning methods will create deeper insights into epileptic networks, going beyond what is visually discernible and addressing the limitations of current diagnostic approaches.
Advancements in deep learning have shown significant potential in various medical applications, particularly in revealing complex patterns within intricate datasets. For instance, recent studies have demonstrated that deep learning models can achieve high accuracy in detecting interictal epileptiform discharges, effectively employ deep learning algorithms to detect epileptic spikes in MEG data and utilize convolutional and recurrent neural networks to detect epileptiform discharges from scalp EEG recordings, illustrating the versatility and effectiveness of deep learning models across different forms of neuroimaging.
With over 200 MEG recordings (from adult patients) and comprehensive clinical data available through the MCW Comprehensive Epilepsy Program, this initiative is uniquely poised to develop and validate a cutting-edge diagnostic tool. The researchers hypothesize that deep learning algorithms will achieve detection accuracy surpassing 90%, with the detected locations correlating strongly with epileptic networks confirmed by invasive EEG.
By enhancing the precision and efficiency of pre-surgical evaluations, this project aims to significantly improve epilepsy management and patient care outcomes in Wisconsin, providing more personalized and effective therapeutic strategies. The forthcoming NIH R21 grant application will seek to expand this research, refining our models for broader clinical application and further advancing health equity by addressing this pressing public health challenge.