A clinical EEG research platform to support progressive model construction

• Published in: 2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) pp. 1 - 6
• Main Authors: Chen, Canbin, Mei, Zhen, Huang, Zhihua
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.11.2022
• Subjects: active learning; Biological system modeling; Brain modeling; clinical medicine; Data models; Electroencephalography; Electroencephalography (EEG); Heuristic algorithms; Machine learning algorithms; machine learning system; Signal processing algorithms; system design and implementation
• DOI: 10.1109/CISP-BMEI56279.2022.9979832

Machine learning algorithms have been widely used in clinical electroencephalography (EEG) studies to help diagnose various diseases and classify patient states. The application of machine learning algorithms in clinical EEG usually requires a large amount of annotated data, and annotating long-term clinical EEG is very expensive. Active learning algorithms can actively query the user for labels, and the number of samples to train a model can often be much lower than the number required in normal supervised learning. We have designed and developed a clinical EEG research platform to support progressive model construction, enabling researchers to focus on clinical EEG data analysis and algorithm design. The platform provides general auxiliary functions for various research tasks, including data format conversion, EEG visualization, annotation, dynamic loading algorithms, and progressive model construction. In our experiments, progressively building models can achieve the same or better performance by labeling only about 18% of the samples as compared to randomly selecting samples for labeling.