Technology

Scanning

Scanning through the EEG signals is an extremely tedious process, taking hours to days for an expert doctor to evaluate and identify seizure signals. FunctionalBrain’s Big Data engine speeds this process up, and selects all signal regions which potentially contain high frequency oscillations i.e. indicators of seizure.

Classifying

The following step is classifying potential seizure signals into confirmed seizure vs non-seizure ones. This requires a careful, lengthy review by experts, which is one of the main contributors to the long waiting periods for EEG related brain surgeries. Machine learning models are being developed as part of FunctionalBrain to increase the accuracy of seizure signal detection.

Machine learning

Instead of using pre-defined mathematical rules, “supervised” machine learning models are built by feeding a large number of EEG signals, each labelled as seizure on non-seizure as appropriate, and letting the algorithm come up with the rules to recognize pathological patterns. Another family of ML algorithms (“unsupervised”) natively categorizes EEG signals into multiple clusters, where signals within the same cluster are similar to one another.

Human skills for optimizing

Our data scientists work together with medical experts to optimize the data transformation and extract the most useful information (features) available from the raw signals. Then, rigorous model selection, optimization and tuning is performed to maximize the performance (maximize precision, minimize false positive hits, ensure model is generalized and fit to make future predictions). Our team’s proven skill set in deploying ML models ensures that the solution can be applied in real-life scenarios.

FunctionalBrain

Technology used