The data challenge
Healthcare data can be notoriously complex. They are often gathered by proprietary software and compiled in siloed databases of largely incompatible systems. Free text reports, notes, images, recordings, and a lack of global standards make a large amount of health data virtually useless in their current state. In addition, there are significant process and technology costs associated with aggregation, cleaning, curating, hosting, analyzing, and protecting the transformation of these raw data records into usable data. This is what we refer to as the data challenge in healthcare.
Collecting datasets of sufficient size and accuracy is a significant challenge, holding back AI development, and will likely continue to be an important issue going forward. The key ingredient to drive this transformation is highly accurate, structured datasets. Structuring and labeling medical data take a combination of training and expertise available only in highly skilled personnel.
Artificial intelligence algorithms on complex data require the combination of raw data and connected labels. The only accepted global standard for the purpose of EEG labeling is SCORE. Applying effective labeling of EEG recordings in clinical production requires dedicated software for the purpose (hiSCORE).
Holberg EEG has collected more than 50 000 de-identified EEG recordings and SCORE reports with high-quality structured data. This database is currently used to develop a decision support system for EEG based on AI -autoSCORE-. autoSCORE aims to reduce EEG reporting variability, improve overall reporting quality, and dramatically reduce cost for the end-user by increasing productivity in the clinical neurophysiology department. This will ultimately lead to improved clinical outcomes and the value of care for the millions suffering from neurological diseases.