Temporal lobe epilepsy (TLE) is a chronic brain disorder where recurrent seizures occur in the temporal lobe. TLE can cause psychological issues, loss of short-term memory etc. and significantly impacts quality of life. Its incidence is reported to range between 0.04 to 0.1% of the global population and therefore is considered to be one of the more prevalent neurological diseases2. TLE can result from multiple causes including, traumatic brain injury, cancer, stroke, infections or scarring in the hippocampus region1. Current treatment paradigms include antiseizure medications, surgery and deep nerve stimulation3, but in some cases, seizures may not be fully managed with available therapies. Consequently, there is an ongoing need to develop improved therapies to manage TLE.
Mouse models of TLE that use pilocarpine or kainic acid to induce seizures have been used to study TLE4 and test therapies, but there are fundamental differences between rodent and human brains in terms of anatomy, physiology and function. Consequently, there is limited translatability from rodent data to human patients. Nonhuman primates (NHPs) are more physiologically relevant models to study TLE due to similarities in structure, function and neurochemical activity with the human brain. Interestingly, epilepsy can develop naturally in NHPs likely due to genetic factors or due to injury or infection, but can also be induced via a wide range of stimuli. Depending on the stimuli that is used, NHPs can develop generalized or focal epilepsy5. Focal epilepsy is induced via alumina gel, pilocarpine, kainic acid or electrical kindling, which uses an implanted stimulation electrode to induce seizures5. Relatively simple NHP epilepsy models have been used widely for the development of anti-seizure therapies. However, an unmet need is the development of new therapies for treatment refractory epilepsy, that need to be evaluated in more complex models that use a combination of stimuli to induce more refractory seizures. One such example, is the combination of pilocarpine and PTZ (pentylenetetrazol), where pilocarpine is used to induce an epilepsy phenotype and low doses of PTZ is used to trigger limbic seizures that are more frequent and severe6. In this model, available therapies reduced seizure intensity and frequency at varying degrees but did not completely suppress the seizures6. This data suggests that complex models that mimic treatment refractory epilepsy could be used to screen for more efficacious therapies.
Epileptic seizures are commonly detected using EEGs (electroencephalograms) where electrodes are positioned around the head to detect changes in brainwave activity. Apart from EEG analysis, diagnostic imaging such as PET, CT scan, MRI etc. are used to identify regions where there are changes in brain activity. Epilepsy patients typically undergo long term EEG scans where data is collected frequently over several days7. Manual analysis of this large dataset can take a long time, is prone to errors and needs to be done by a trained reader or experienced neurologists. Therefore, this type of analysis can be a significant bottleneck for the timely diagnosis and management of epilepsy. However, artificial intelligence (AI) can be a valuable aid for this analysis and can help reduce the error rate and time. In 2017, the Cleveland Clinic partnered with Google Inc. to develop a deep learning neural network to analyze a huge dataset (20 terabytes) from epilepsy patients7. The collaboration resulted in the development of a temporal graph convolutional network (TGCN) from the EEG data of 995 patients. This model combines spatial data over a set time period7 and showed impressive sensitivity and specificity7. Recently, a group at University College London developed an AI algorithm to identify areas of abnormal brain dysplasia that could lead to epileptic seizures using MRI data from 538 patients8. The algorithm was able to detect brain abnormalities in about 67% of the cases.
It is clear that AI is being actively used as a tool to identify and monitor epileptic seizures in human patients. However, it is important to reverse translate these AI algorithms to NHP epilepsy models so that AI and machine learning platforms can aid in the preclinical development of new therapies for treatment resistant epilepsy.