Competition Title: Classification of Harmful Brain Activity

Subtitle: Detect seizures and related harmful brain activity in electroencephalography data

Goal of the Competition: 

The goal of this competition is to detect and classify seizures and other types of harmful brain activity in electroencephalography (EEG) signals. Your work will help doctors and researchers detect seizures in patients with epilepsy and brain injuries and help companies developing new drugs to treat and prevent seizures. Examples of the types of brain activity you will be aiming to classify are shown in Figure 1. 

(Add Figure 01)

Context:

EEG monitoring is used in the development of drugs designed to treat or prevent seizures, and for monitoring the safety of new drugs under development that might place patients at risk for seizures. Such EEG data is typically recorded continuously for multiple hours, and subsequently interpreted by manual review of the EEG data by experts trained in neurology with subspecialized training in epilepsy. This approach is not only expensive and time consuming but is also prone to error because of problems with inter-rater reliability and data fatigue.

A machine learning model that can accurately detect and classify seizures and other harmful patterns of brain activity automatically at a level of accuracy matching or exceeding that of typical experts could make the interpretation of clinical data and clinical trial data faster, less expensive, and more reliable.

Drs. Jing, Ge, Sun and Westover have developed the world's largest set of EEG data labeled by multiple clinical experts and created a proof-of-concept machine learning model for accomplishing this task (https://bdsp.io/content/bdsp-sparcnet/1.1/ ). This competition aims to engage the machine learning community to develop a definitive solution to the problem of detecting and classifying harmful patterns of brain activity in EEG recordings. The competition is sponsored by and will be run in collaboration with ModelShareAI, which specializes in enabling AI computing contests and AI education, Persyst, which develops EEG analysis software, and Jazz Pharmaceuticals, which specializes in developing drugs to treat epilepsy.

We’re excited to delve into the responses we get to this challenge. If successful, the algorithms that you develop by participating in this contest may help neurology researchers better understand which types of seizures injure the brain and how, and help doctors to use the algorithms to detect and treat seizures more quickly. Your algorithms will also help companies to make safer and more effective drugs for epilepsy and other neurologic diseases. As a result, new drugs could be discovered to treat the millions of people with these leading causes of death and disability.

To participate in the contest, create an account on: modelshare.ai 

The Brain Data Science Platform (BDSP) Seizure Dataset is the worlds large collection of EEG recordings annotated by multiple clinical experts. The dataset was developed as part of an NIH R01 grant (R01NS107291). The dataset has not-yet been publicly released, which is ideal for the present competition.

The dataset includes 6,095 noninvasive EEGs recordings from 2,711 distinct patients. The dataset contains 71,982 ten-second-long EEG segments that have been independently annotated by 20 fellowship-trained neurophysiologists. This intensive labeling by multiple clinical experts allows robust evaluation of the degree to which machine learning algorithms match expert performance.

Competition details are provided here.

The data is also available here.