Machine Intelligence-Based Epileptic Seizure Forecasting

Epilepsy is one of the most common neurological disorders globally, and the decrease in quality of life associated with it includes – among other things – fear and uncertainty over when the next seizure would manifest itself. The most common way to treat epilepsy is by using antiepileptic drugs; however, around 30% of all patients develop refractory epilepsy, where medication fails to control seizures, and patients have to resort to surgical resection of epileptogenic zones. While manual techniques exist to detect epileptic seizures, and come up with the appropriate regiment of antiepileptic drugs, they are generally limited by the skill of the human operator and can be applied only to a particular application. Arguably, a better approach is to use machine intelligence to identify patterns in data unseen to the human eye and perform identification of seizure states, and medicine regiments in an automated objective manner. In this chapter, we will discuss such machine learning algorithms. We will explore the most widely used algorithms and their variations – both in the context of seizure prediction and detection (arguably the most widely used application of machine intelligence in epilepsy), as well as in other applications, such as antiepileptic drug efficacy. We will also talk about common techniques of feature extraction – particularly focusing on wavelet phase coherence and cross-frequency coupling. While much of work has been done to improve current machine learning algorithms in the context of epilepsy, challenges still remain to be solved, and potential future directions for machine intelligence applications in epilepsy are discussed at the end of the chapter.