Projects

AI techniques have been successfully applied to a variety of biomedical signal (image) processing fields such as high-quality-imaging, reconstruction, analysis, classification, disease detection, multi-modal fusion, etc., due to its state-of-the-art performance. Our team exploits and develops top-notch AI methods based on medical data, medical instruments, clinical knowledge, and pathological characteristics for strengthening healthcare. To be successful, we access to a vast array of imaging technologies, computer science, medical and data science experts, and an environment with low disciplinary barriers. Also, we collaborate with PNU Yangsan Hospital and Pusan Medical Center, equipped with advanced facilities including MRI scanners, PET/CT scanners, and ultrasound systems, to obtain and process data. School of Medicine provides the medical knowledge base and patient access required.

Our research extends ultrasound (US) and photoacoustic techniques to provide sensitive functional images using current commercial instruments or developing systems. US techniques currently provide greater depth of penetration in widely available and real-time devices. Photoacoustic techniques are a promising approach to bring the molecular sensitivity of optical contrast mechanisms into practical US systems used in clinic. Their advance restores the low-cost and high-safety aspects. The fundamental contribution of this research is to increase the sensitivity and specificity to detect abnormal lesions. This research focuses on implementing measurement systems most effective at achieving any clinical object, and developing novel signal/image processing and machine learning techniques to isolate signal components specific to the clinical task.

Near-infrared spectroscopy (NIRS) is also an emerging optical modality to access changes in hemoglobin concentration non-invasively in brain tissue. The safe, portable and affordable procedure allows a wide scope of applications including cognitive neuroscience, brain activation research, and clinical monitoring of neurological diseases. However, the biggest technical challenge arises from unwanted data variations due to voluntary and involuntary motions. The consequent artifacts incredibly disturb the detection of the blood-dependent signals. The research aims to develop a deep-learning technique for estimating hemodynamic responses. Brain imaging with motion compensation is expected to have broad clinical and scientific applicability. Our plan is to apply the method to the clinical NIRS data acquired for the study of attention deficit hyperactive disorder (ADHD) children. We expect that results help us to successfully visualize the hypoactivation pattern during a task in ADHD children compared with typically developing control children at a group level. Collaboration groups are the Korea Basic Science Institute and Dan’s Lab at Chuo University.