https://ir.ymlib.yonsei.ac.kr/handle/22282913/168940 2026-04-05T20:27:48Z https://ir.ymlib.yonsei.ac.kr/handle/22282913/211233 Title: Classification of twinkling artifacts and blood flow for in vivo detection of breast microcalcifications Authors: Kang, Jinbum; Park, Seongjun; Lee, Eonho; Cho, Hyunwoo; Kim, Kangsik; Kim, Min Jung; Yoo, Yangmo Abstract: While mammography is the standard modality for detecting microcalcifications (MCs), their real-time detection with ultrasound imaging can be invaluable, particularly for guiding biopsies. Ultrasound twinkling artifact (TA) imaging allows the sensitive distinction of MCs from background breast tissue; however, it may also be confounded with blood flow in Doppler mode during in vivo scanning. In this paper, we propose a new MC imaging method that classifies TA and blood flow signals to enable in vivo detection of breast MCs. Based on the signal characteristics of TA and blood flow, two optimal features (i.e., mean frequency and spectrum bandwidth) are extracted and used to train a machine learning classifier. To train the classification model, tissue-mimicking and chicken breast phantom containing normal wire (285 mu m in diameter), MC wire (300 mu m in diameter) and micro-vessel tube (1 mm in diameter) were fabricated, and training and validation datasets were acquired under varying flow velocities and pulse repetition frequencies (PRFs). Among the four classifiers, i.e., k-nearest neighbors (KNN), support vector machine (SVM), na & iuml;ve Bayes and quadratic discriminant, trained with the two optimal features, the SVM achieved the highest accuracy (95.25 %), whereas the remaining models also exhibited strong performance with accuracies exceeding 92 %. The trained SVM model was then validated on a chicken breast MC phantom and in vivo human breast data, and they showed good agreement with color Doppler imaging. The feasibility study demonstrated that the proposed classification approach may enable effective in vivo detection and improve diagnostic accuracy, especially in cases with complex flow patterns in breast lesions. 2026-07-01T00:00:00Z https://ir.ymlib.yonsei.ac.kr/handle/22282913/211295 Title: HP-GAN: Harnessing pretrained networks for GAN improvement with FakeTwins and discriminator consistency Authors: Son, Geonhui; Lee, Jeong Ryong; Hwang, Dosik Abstract: Generative Adversarial Networks (GANs) have made significant progress in enhancing the quality of image synthesis. Recent methods frequently leverage pretrained networks to calculate perceptual losses or utilize pretrained feature spaces. In this paper, we extend the capabilities of pretrained networks by incorporating innovative self-supervised learning techniques and enforcing consistency between discriminators during GAN training. Our proposed method, named HP-GAN, effectively exploits neural network priors through two primary strategies: FakeTwins and discriminator consistency. FakeTwins leverages pretrained networks as encoders to compute a self-supervised loss and applies this through the generated images to train the generator, thereby enabling the generation of more diverse and high quality images. Additionally, we introduce a consistency mechanism between discriminators that evaluate feature maps extracted from Convolutional Neural Network (CNN) and Vision Transformer (ViT) feature networks. Discriminator consistency promotes coherent learning among discriminators and enhances training robustness by aligning their assessments of image quality. Our extensive evaluation across seventeen datasets-including scenarios with large, small, and limited data, and covering a variety of image domains-demonstrates that HP-GAN consistently outperforms current state-of-the-art methods in terms of Fr & eacute;chet Inception Distance (FID), achieving significant improvements in image diversity and quality. Code is available at: https://github.com/higun2/HP-GAN. 2026-07-01T00:00:00Z https://ir.ymlib.yonsei.ac.kr/handle/22282913/211609 Title: Web-based discharge education program for caregivers of children with epilepsy: A feasibility study Authors: Lee, Hyunjie; Oh, Eui Geum; Lee, Seung Eun; Kang, Hoon-Chul; Shin, Hae-kyung; Cha, Yerin; Choi, Eun Kyoung Abstract: Purpose: This study aimed to develop and evaluate the feasibility of a theory-driven, web-based discharge education program designed to enhance self-management among caregivers of children with epilepsy. Methods: Following the Analyze, Design, Develop, Implement, and Evaluate (ADDIE) model, educational needs were assessed through an integrative literature review and focus group interviews with caregivers and pediatric nurses. The program was structured using the Include, Discuss, Educate, Assess, and Listen (IDEAL) discharge planning framework, and individual and family self-management theory. Results: The key features included self-assessment tools, pre-discharge questions, multimedia materials, health diaries, and Q&A boards. Content validity was reviewed by experts, usability tested with caregivers and professionals, and clinical feasibility examined using a one-group pretest-posttest design involving 13 caregivers. Participants reported high satisfaction with the clarity and accessibility of the content, with strongest engagement in the self-assessment and pre-discharge sections. Although changes in discharge readiness and selfmanagement scores were not statistically significant, moderate-to-large effect sizes suggested an increasing trend in practical improvement. Caregivers also reported improved confidence, medications awareness, and follow-up adherence. Feedback supported mobile-based delivery and highlighted the need for interactive and personalized features. Conclusions: The program is a feasible and accepted intervention that incorporates self-assessment into discharge education and leverages web-based delivery to support family centered care. Practice implications: The program addresses existing gaps in pediatric epilepsy education and shows potential for broader implementation. Integrating self-assessment and digital platforms into discharge education may enhance caregivers' preparedness and support sustainable self-management. (c) 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies. 2026-05-01T00:00:00Z https://ir.ymlib.yonsei.ac.kr/handle/22282913/211104 Title: FeDi: Feature disentanglement for self-supervised learning Authors: Lee, Jeong Ryong; Son, Geonhui; Hwang, Dosik Abstract: Self-supervised learning (SSL) has revolutionized the field of deep learning by enabling the extraction of meaningful representations from unlabeled data. In this work, we introduce FeDi, a novel SSL method that leverages feature disentanglement to enhance the quality and robustness of learned representations. FeDi maximizes the lower bound on mutual information between representation vectors across batch dimensions, effectively disentangling features and preventing representation collapse. Our proposed method serves as a hardness-aware loss function that automatically balances alignment and disentanglement terms, effectively managing the challenges of disentangling high-dimensional representations. Our extensive experiments demonstrate that FeDi consistently outperforms state-of-the-art SSL methods across a variety of tasks, including image classification, object detection, and segmentation. Code is available at: https://github.com/mongeoroo/fedi. 2026-04-01T00:00:00Z