https://scholar.dgist.ac.kr/handle/20.500.11750/128 2026-04-04T20:35:29Z https://scholar.dgist.ac.kr/handle/20.500.11750/58937 Title: Diagnosis of Incomplete Kawasaki Disease using Deep learning Techniques with Ultrasound Images of Coronary Artery Lesions Author(s): Lee, Haeyun; Lee, Moon Hwan; Youngmin, Lucy; Eun, Yongsoon; Hwang, Jae Youn Abstract: Kawasaki disease (KD) is the most common cause of acquired heart disease in young children and can lead to sudden death. Incomplete KD lacks clinical characteristics of KD and is thus difficult to distinguish from other diseases presenting similar symptoms. Although ultrasound imaging is useful to identify one of the most fatal complications, coronary aneurysms, the diagnosis of incomplete KD is still difficult due to its similar symptoms to other diseases. We here demonstrated the feasibility of the deep learning algorithms for the diagnosis of incomplete KD. Various deep learning networks were trained, and their accuracy was compared. Although the accuracy is lower than the experienced specialist, the experimental results suggest that deep learning algorithms may assist clinicians to diagnose KD. © ICA 2022.All rights reserved 2022-10-26T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/58936 Title: Bladder volume estimation deep learning algorithm using depth dependent coefficients of ultrasound signals Author(s): Kang, Minji; Lee, Moon Hwan; Hwang, Jae Youn Abstract: Bladder volume estimation in patients with dysuria is performed through ultrasound imaging. Estimation of bladder volume with bladder ultrasound images differs from the actual volume by an average of 18% when the bladder is assumed to have a spherical shape without considering the difference in a bladder shape along a bladder volume. To overcome this issue, we demonstrate a deep learning-based bladder volume estimation network that is capable of reducing volume estimation errors as the shape of the bladder changes. The proposed network synthesizes a few scanline images into an ultrasound image with a large number of scanlines using the combination of GAN(Pix2Pix) and U-Net architectures. The network shows an accuracy of 93% in terms of IoU, demonstrating the applicability of the bladder ultrasound wearable system for the segmentation of bladder regions with a few scanlines. © ICA 2022.All rights reserved 2022-10-26T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/58935 Title: Measurement of single-cell viscoelastic properties using acoustic tweezers Author(s): Park, Hee Yeon; Youn, Sangyeon; Kim, Jae Seong; Lee, Moon Hwan; Hwang, Jae Youn Abstract: The mechanical property of a cell is a key indicator that shows the pathological characteristics of cells. Cell viscoelasticity is one of the mechanical properties of the cell that exhibits both viscous and elastic characteristics when cell deformation occurs, and a changing acoustic pressure is required when analyzing them. In the measurement of the mechanical properties of cells, acoustic tweezers have been utilized in a non-invasive manner since they have several advantages such as inducing less cell damage and generating high-trapping force compared to optical tweezers. To quantify the viscoelasticity of a cell, a press-focused single-element ultrasound transducer was fabricated. It was used to trap a target cell and then change the deformation by varying trapping forces. The deformation of breast cancer cells at different trapping forces was recorded in a high-speed camera for quantification of cell deformability. Acoustic pressure at different pressures was applied to the cell for quantification of deformation and comparison of metastability. Also, it was found that the amplitude modulated force resulted in a phase delay between applied forces and cell deformation while cell trapping, suggesting that by analyzing the phase delay, the viscosity of the trapped cell can be quantified. © ICA 2022.All rights reserved 2022-10-26T15:00:00Z https://scholar.dgist.ac.kr/handle/20.500.11750/58654 Title: Connectome Mapping: Shape-Memory Network via Interpretation of Contextual Semantic Information Author(s): Lee, Kyungsu; Lee, Haeyun; Hwang, Jae Youn Abstract: Contextual semantic information plays a pivotal role in the brain's visual interpretation of the surrounding environment. When processing visual information, electrical signals within synapses facilitate the dynamic activation and deactivation of synaptic connections, guided by the contextual semantic information associated with different objects. In the realm of Artificial Intelligence (AI), neural networks have emerged as powerful tools to emulate complex signaling systems, enabling tasks such as classification and segmentation by understanding visual information. However, conventional neural networks have limitations in simulating the conditional activation and deactivation of synapses, collectively known as the connectome, a comprehensive map of neural connections in the brain. Additionally, the pixel-wise inference mechanism of conventional neural networks failed to account for the explicit utilization of contextual semantic information in the prediction process. To overcome these limitations, we developed a novel neural network, dubbed the Shape Memory Network (SMN), which excels in two key areas: (1) faithfully emulating the intricate mechanism of the brain's connectome, and (2) explicitly incorporating contextual semantic information during the inference process. The SMN memorizes the structure suitable for contextual semantic information and leverages this structure at the inference phase. The structural transformation emulates the conditional activation and deactivation of synaptic connections within the connectome. Rigorous experimentation carried out across a range of semantic segmentation benchmarks demonstrated the outstanding performance of the SMN, highlighting its superiority and effectiveness. Furthermore, our pioneering network on connectome emulation reveals the immense potential of the SMN for next-generation neural networks. © 2025 13th International Conference on Learning Representations, ICLR 2025. All rights reserved. 2025-04-24T15:00:00Z