For how long do the patients have to wear the device to collect the data?
Amba: It depends on the use case. The patients could wear it for a short period of time. For example, if the goal is to collect brain data for epilepsy, the patient can wear it for ten to fifteen minutes per day for a brief period of time before or after a seizure episode. In other use cases such as mental health or depression, the physicians or psychologists is expected to prescribe the duration of wearing the device, depending on the amount of time they intend to collect the data.
The goal of Evercot AI is to provide a medical grade wearable device that people can carry along with them to capture and monitor their brain waves at any time or location. This data can be automatically and wirelessly sent to the physician’s medical center or to the cloud. Or, if the data is collected for personal reasons, it could be used for whatever the patient want to do with the data.
Are you planning to further develop your products?
Amba: Yes, there are two points we are constantly working on. First, unlike in a hospital setting where EEG data is collected when the patient is sitting, NioWear EEG headset is expected to acquire medical grade EEG data while the patient is performing her day to day activities. Such activities usually involve motion, which would lead to more noise and movement artifacts during mobile EEG data collection. Based on this, we are continuously improving our advanced signal post-processing approaches to remove unwanted movement and environmental artifacts — to ensure that good medical grade mobile EEG data is acquired.
Secondly, we are constantly building novel predictive machine learning algorithms that can diagnose, predict or manage brain diseases from EEG data. Our brain predictive algorithmic solutions are either deployed and delivered in a companion app or could also be accessed through the Greey Matter platform, which I alluded earlier. Lately we developed a machine learning algorithm for the epilepsy use case. And the next project we are working on is to develop algorithms for e.g., depression, mental health and Parkinson’s disease.
How exact does your AI work when it comes to detecting brain diseases?
Amba: There are two types of brain data. Sensor brain data (e.g., EEG data) and medical image brain data (e.g., MRI, CT and X-Ray scans). The way our AI works on detecting brain disease can be better articulated by explaining how the algorithm handles brain data. The approach we use in crafting a predictive brain disease AI solution slightly differs if it’s a sensor brain data or a medical image brain data.
Generally, at a higher level, the brain data is first pre-processed. EEG pre-processing involves the removal of noise and motion artifacts. During medical image data-pre-processing, techniques such as Skull Stripping is employed to separate and extract brain tissue from skull. This is followed by registration to ensure that all volumes are align in the same orientation. The medical images are then resampled, and intensity normalization is performed on the images.
Once the brain data has been pre-processed, important features or signatures are extracted from the data. In the case of medical images, we utilize Deep Learning and some Reinforcement Learning to build predictive models for a given brain disease. On the other hand, with sensor brain data (i.e., EEG) we use time series data mining techniques, and a broader group of supervised, unsupervised and outlier detection machine learning techniques to build the brain predictive models. We partition the historical brain data into training and test dataset. The brain disease predictive model is trained on the training set and the performance of the model is tested on the test dataset.
Once the model is on the production system, when new unseen patient data is uploaded to perform diagnostics, the model would provide a prediction result, which is usually diagnostic probabilities if a patient has a specific brain disease. Today, most of the medical EEG data is collected on the hospital premise. NioWear brain wearable provides an opportunity to gather EEG data remotely, thus accelerating telemedicine for brain related diseases. We believe this is going to be the future of brain and general diagnostics in health care. Last year, we started collecting medical grade mobile brain EEG data with our device outside the hospital. The more the historical mobile EEG data we acquire, the better the performance of the predictive AI for a given brain disease. An increased performance of the brain disease AI algorithm is beneficial to patients and elevate many manual tasks burden from physicians. Evercot AI is interested in data-sharing partnerships with physicians, medical centers and medical research institutes who aspire to utilize AI and Big Data to improve early disease diagnostics or increase the efficiency of health care workflows.
You gave a lecture about Evercot.AI at the virtual.MEDICA 2020 – how did you like it? Do you think a virtual trade fair can supplement a normal trade fair?
Amba: Based on the short amount of time that was used to set up the virtual trade fair, it was excellent. It was hard to determine the difference between a physical and a virtual trade fair. I think it was a great experience and it was also very comfortable to visit any presentation with just one click of the mouse. But I think, in most cases a physical trade fair is still very relevant because a virtual trade fair can’t substitute the human face-to-face interaction. But maybe, in the future a hybrid trade fair could be a solution: We could be at a physical trade fair but still have the computer where we are capable of virtually visiting some presentations.
