Associate director of Bayesian learning team
Heung-il Suk (Korea Univ.)
(1) Brain functional network monitoring
(2) Brain disease diagnosis/prognosis
Participant professors of Bayesian learning team
Jaeho Han (Korea Univ.)
(1) Retinal image segmentation
(2) Bio signal analysis
Brain functional network monitoring
As the population becomes older worldwide, neurodegenerative brain disorders, e.g., Alzheimer’s disease, have been considered as one of the major social issues. In our group, we aim to develop a brain monitoring system that estimates brain functional networks from resting-state functional Magnetic Resonance Imaging (fMRI) data via machine learning techniques. Specifically, we utilize a deep architecture to discover latent non-linear associations among brain regions, which are spatially segregated but functionally integrated, and analyze functional characteristics of a brain with computational models such as dynamic state-space model, sparse regression model, etc. With the development of a brain network monitoring system, it can be used to predict the progression of a brain disease in the clinic.
Brain disease diagnosis/prognosis
For an expert system of brain disorder diagnosis or prognosis, we aim to develop novel machine learning methods that can integrate multiple neuroimaging modalities, e.g., MRI, PET, fMRI, and genetic data, so that we maximally utilize the complimentary information inherent in different modalities. The brain disorders of our interests include Alzheimer's Disease (AD) and its prodromal stage Mild Cognitive Impairment (MCI), Autism Spectral Disorder (ASD), and schizophrenia. Upon successful development of the system, it can be used to support clinical decisions, and thus to enhance diagnostic accuracy.
Retinal image segmentation
In this research center, we are developing a novel automated retinal image segmentation algorithm for diagnosis of retinal diseases. Manual segmentation for retinal images is not only extremely time consuming for clinical use but also demanding well-trained expert graders. Moreover, with the developments in acquisition speed and image quality, the amount of data for segmentation has been increased. Therefore, automated segmentation for retinal images is regarded as a core technique in order to improve a productivity. Here, machine learning can guide us to show a powerful performance for each retinal disease which shows a specific pattern. Each of specific patterns can be acquired from quite a number of retinal image samples and the parameters of segmentation algorithm can be tuned through the machine such as support vector machine (SVM), Gaussian mixture model (GMM) clustering, Convolutional neural networks (CNN), and so on. Consequently, the automated segmentation algorithm based on the machine learning achieves greater performance than other segmentation approaches like graph theory and peak detection.
Bio signal analysis
In this research center, we are developing efficient methods for analyzing bio signal such as electroencephalogram (EEG), electrocardiography (ECG), and electromyography (EMG) signals. Here, machine learning approaches are applied as one of perfect candidates for compensating the high variability in EEG, ECG, and EMG. Especially, we focus on change of bio signals when evoking brain neural activities in order to control animal’s behaviors. Moreover, animal’s minute vibrations play as a noise, so machine learning is regarded as an essential part when we are extracting the significant bio signals.