Yuan Lin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Yuan Lin has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Debasish Ghose
Kristiania University of Applied Sciences, Bergen, Norway
5 shared publications - Juan Liu
Pen-Tung Sah Institute of Micro-Nano Science and Technology, Xiamen University, Xiamen, China
4 shared publications - Xiangrong Liu
School of Informatics, Xiamen University, Xiamen, China
3 shared publications - Bithi Banik
Kristiania University of Applied Sciences, Bergen, Norway
2 shared publications - Andrii Shalaginov
School of Technology and Innovation, Kristiania University of Applied Sciences, Kirkegata 24, 0153 Oslo, Norway
1 shared publication - Bjarte Nerland
Kristiania University of Applied Sciences, Bergen, Norway
1 shared publication
Latest Publications
- Open Access
- 1 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computing for Healthcare
From Challenges to Solutions: A Systematic Review of Multimodal LLMs in Healthcare and Architectural Recommendations with RAG
With the evolution of intelligent systems, there is an increasing need to incorporate multimodal inputs, such as medical images, clinical text, structured records, and physiological signals, to facilitate precise and contextually informed clinical decision-making. To cope with multimodal healthcare data, challenges like computational complexity, data heterogeneity, and privacy concerns are rising regardless of advancements in large language model (LLM) capabilities. Applying the PRISMA methodology, this systematic review examined the present status of research on multimodal large language models (MLLMs) in healthcare, published between 2019 and 2024. A total of 164 studies were reviewed, focusing on the types of multimodal data utilized, the fusion techniques and LLM architectures applied, together with the key challenges and emerging opportunities in multimodal processing. The analysis reveals that many existing approaches still lack a comprehensive understanding of MLLM architectures and their role in clinical decision-making, and they also often lack robust privacy-preserving mechanisms and sufficient validation in real-world clinical workflows. Based on these observations, we recommend a customizable framework consisting of interoperable components that can be configured to meet the needs of specific clinical use cases. Instead of proposing a new algorithm or rigid pipeline, this recommendation incorporates a retrieval-augmented generation (RAG) framework coupled with an LLM that can integrate multimodal healthcare data. A central focus of this design is the integration of privacy-preserving retrieval strategies with RAG workflows to ensure secure handling of multimodal data, which highlights the need to examine the role of multimodal data encoding and storage, encrypted retrieval, reranking of retrieved data, multimodal prompt engineering, and secure content generation. Together, these components enable the integration of RAG with MLLMs in a way that supports privacy-aware, explainable, and clinically grounded decision support. The insights presented aim to guide future efforts to develop secure, explainable, and clinically integrated intelligent systems that take advantage of RAG and multimodal LLM.
- 1 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computing for Healthcare
Open AccessFrom Challenges to Solutions: A Systematic Review of Multimodal LLMs in Healthcare and Architectural Recommendations with RAG
With the evolution of intelligent systems, there is an increasing need to incorporate multimodal inputs, such as medical images, clinical text, structured records, and physiological signals, to facilitate precise and contextually informed clinical decision-making. To cope with multimodal healthcare data, challenges like computational complexity, data heterogeneity, and privacy concerns are rising regardless of advancements in large language model (LLM) capabilities. Applying the PRISMA methodology, this systematic review examined the present status of research on multimodal large language models (MLLMs) in healthcare, published between 2019 and 2024. A total of 164 studies were reviewed, focusing on the types of multimodal data utilized, the fusion techniques and LLM architectures applied, together with the key challenges and emerging opportunities in multimodal processing. The analysis reveals that many existing approaches still lack a comprehensive understanding of MLLM architectures and their role in clinical decision-making, and they also often lack robust privacy-preserving mechanisms and sufficient validation in real-world clinical workflows. Based on these observations, we recommend a customizable framework consisting of interoperable components that can be configured to meet the needs of specific clinical use cases. Instead of proposing a new algorithm or rigid pipeline, this recommendation incorporates a retrieval-augmented generation (RAG) framework coupled with an LLM that can integrate multimodal healthcare data. A central focus of this design is the integration of privacy-preserving retrieval strategies with RAG workflows to ensure secure handling of multimodal data, which highlights the need to examine the role of multimodal data encoding and storage, encrypted retrieval, reranking of retrieved data, multimodal prompt engineering, and secure content generation. Together, these components enable the integration of RAG with MLLMs in a way that supports privacy-aware, explainable, and clinically grounded decision support. The insights presented aim to guide future efforts to develop secure, explainable, and clinically integrated intelligent systems that take advantage of RAG and multimodal LLM.
- Open Access
- 30 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
FedBalance: An Adaptive Cross-Silo Federated Learning Approach for Stress Detection under non-IID Data Distributions
- 30 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessFedBalance: An Adaptive Cross-Silo Federated Learning Approach for Stress Detection under non-IID Data Distributions
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- p. 1
FedSmartCare: Design and Implementation of a Federated Learning Enabled Vital Sign Monitoring System
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- p. 1
Open AccessFedSmartCare: Design and Implementation of a Federated Learning Enabled Vital Sign Monitoring System
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (2) , 2078-2088
IAHED: An Imbalance-Adaptive Hybrid Encoder–Decoder Framework for Early Detection of ICU-Acquired Multidrug-Resistant Bacteria
The rising incidence of Intensive Care Unit-acquired Multidrug-Resistant Bacteria (ICU-acquired MDRB) presents significant public health challenges. Traditional diagnostic methods are time-consuming, while advanced molecular diagnostics remain expensive and not widely adopted in clinical practice. Although electronic health records (EHR) offer potential for developing effective diagnostic models, extreme data imbalance and complexity inherent in ICU settings limit conventional approaches. We propose an Imbalance-Adaptive Hybrid Encoder-Decoder (IAHED) framework for early ICU-acquired MDRB detection, evaluated on 25,864 ICU stays from the MIMIC-IV database. We analyzed two observation cohorts: 6,358 ICU stays (76 MDRB cases) for 7-day observation and 2,160 ICU stays (63 MDRB cases) for 14-day observation, with class imbalance ratios of 0.012 and 0.029, respectively. Cross-domain validation was conducted using the eICU Collaborative Research Database to evaluate generalizability across different healthcare institutions, with observation cohorts of 2,847 ICU stays (55 MDRB cases) for 7-day observation and 1,280 ICU stays (34 MDRB cases) for 14-day observation. IAHED integrates dynamic and static patient data through a dual-pathway architecture, employing Givens rotations for data augmentation and a composite loss function incorporating contrastive learning to enhance minority class detection. Model interpretability was ensured through SHAP (SHapley Additive exPlanations) analysis to identify key clinical predictors and support clinical decision-making. Experimental results demonstrate superior performance over established machine learning and deep learning methods, achieving AUROC of 0.950-0.963 and AUPRC of 0.857-0.878, with cross-domain validation on eICU demonstrating transferability across both observation windows (AUROC 0.896-0.908), with particularly robust performance in highly imbalanced scenarios. These findings highlight IAHED’s potential for improving diagnostic accuracy and supporting timely clinical decision-making in ICU settings.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (2) , 2078-2088
IAHED: An Imbalance-Adaptive Hybrid Encoder–Decoder Framework for Early Detection of ICU-Acquired Multidrug-Resistant Bacteria
The rising incidence of Intensive Care Unit-acquired Multidrug-Resistant Bacteria (ICU-acquired MDRB) presents significant public health challenges. Traditional diagnostic methods are time-consuming, while advanced molecular diagnostics remain expensive and not widely adopted in clinical practice. Although electronic health records (EHR) offer potential for developing effective diagnostic models, extreme data imbalance and complexity inherent in ICU settings limit conventional approaches. We propose an Imbalance-Adaptive Hybrid Encoder-Decoder (IAHED) framework for early ICU-acquired MDRB detection, evaluated on 25,864 ICU stays from the MIMIC-IV database. We analyzed two observation cohorts: 6,358 ICU stays (76 MDRB cases) for 7-day observation and 2,160 ICU stays (63 MDRB cases) for 14-day observation, with class imbalance ratios of 0.012 and 0.029, respectively. Cross-domain validation was conducted using the eICU Collaborative Research Database to evaluate generalizability across different healthcare institutions, with observation cohorts of 2,847 ICU stays (55 MDRB cases) for 7-day observation and 1,280 ICU stays (34 MDRB cases) for 14-day observation. IAHED integrates dynamic and static patient data through a dual-pathway architecture, employing Givens rotations for data augmentation and a composite loss function incorporating contrastive learning to enhance minority class detection. Model interpretability was ensured through SHAP (SHapley Additive exPlanations) analysis to identify key clinical predictors and support clinical decision-making. Experimental results demonstrate superior performance over established machine learning and deep learning methods, achieving AUROC of 0.950-0.963 and AUPRC of 0.857-0.878, with cross-domain validation on eICU demonstrating transferability across both observation windows (AUROC 0.896-0.908), with particularly robust performance in highly imbalanced scenarios. These findings highlight IAHED’s potential for improving diagnostic accuracy and supporting timely clinical decision-making in ICU settings.
- Open Access
- 12 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
ChiralRetro: Chirality-Sensitive Encoding and Enhanced Pathway Strategies in Retrosynthesis
- 12 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Open AccessChiralRetro: Chirality-Sensitive Encoding and Enhanced Pathway Strategies in Retrosynthesis
- Open Access
- 18 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 59807-59817
FishIR: Identifying Pufferfish Individual Based on Deep Learning and Face Recognition
Pufferfish, globally recognized for its distinctive delicacy, carries high culinary value. However, it is also notorious for the lethal toxicity, and there is a great demand for traceability measures in the commercial trade of pufferfish to assure safety and accountability. This research introduces a novel deep learning approach, utilizing facial recognition techniques, to identify pufferfish individuals. This method specifically leverages distinctive back skin texture patterns as key biological traits. Our initial step involved assembling a collection of annotated and augmented images of Takifugu bimaculatus, a species of pufferfish native to East China Sea, which is accessible upon request. We then extensively investigated fundamental components of Deep Face Recognition (deep FR) systems, focusing on segmentation and extraction models, and assessed their effectiveness in identifying pufferfish. Following this, we developed FishIR (Fish Individual Recognition), a framework to identify pufferfish individuals that consists of four deep FR stages while incorporating enhanced segmentation and feature extraction techniques. Experimental results show that this framework successfully captures unique representations of individual pufferfish, as verified by the high accuracy achieved in recognition tasks.
- 18 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 59807-59817
Open AccessFishIR: Identifying Pufferfish Individual Based on Deep Learning and Face Recognition
Pufferfish, globally recognized for its distinctive delicacy, carries high culinary value. However, it is also notorious for the lethal toxicity, and there is a great demand for traceability measures in the commercial trade of pufferfish to assure safety and accountability. This research introduces a novel deep learning approach, utilizing facial recognition techniques, to identify pufferfish individuals. This method specifically leverages distinctive back skin texture patterns as key biological traits. Our initial step involved assembling a collection of annotated and augmented images of Takifugu bimaculatus, a species of pufferfish native to East China Sea, which is accessible upon request. We then extensively investigated fundamental components of Deep Face Recognition (deep FR) systems, focusing on segmentation and extraction models, and assessed their effectiveness in identifying pufferfish. Following this, we developed FishIR (Fish Individual Recognition), a framework to identify pufferfish individuals that consists of four deep FR stages while incorporating enhanced segmentation and feature extraction techniques. Experimental results show that this framework successfully captures unique representations of individual pufferfish, as verified by the high accuracy achieved in recognition tasks.
- 29 March 2023
- Published by Springer Nature
- p. 515-526
Fast Accurate Fish Recognition with Deep Learning Based on a Domain-Specific Large-Scale Fish Dataset
- 29 March 2023
- Published by Springer Nature
- p. 515-526
Fast Accurate Fish Recognition with Deep Learning Based on a Domain-Specific Large-Scale Fish Dataset
- 30 December 2013
- Published by American Physical Society (APS) in Physical Review E
- Vol. 88 (6) , 062724
Collective behavior and predation success in a predator-prey model inspired by hunting bats
We establish an agent-based model to study the impact of prey behavior on the hunting success of predators. The predators and prey are modeled as self-propelled particles moving in a three-dimensional domain and subject to specific sensing abilities and behavioral rules inspired by bat hunting. The predators randomly search for prey. The prey either align velocity directions with peers, defined as “interacting” prey, or swarm “independently” of peer presence; both types of prey are subject to additive noise. In a simulation study, we find that interacting prey using low noise have the maximum predation avoidance because they form localized large groups, while they suffer high predation as noise increases due to the formation of broadly dispersed small groups. Independent prey, which are likely to be uniformly distributed in the domain, have higher predation risk under a low noise regime as they traverse larger spatial extents. These effects are enhanced in large prey populations, which exhibit more ordered collective behavior or more uniform spatial distribution as they are interacting or independent, respectively.
- 30 December 2013
- Published by American Physical Society (APS) in Physical Review E
- Vol. 88 (6) , 062724
Collective behavior and predation success in a predator-prey model inspired by hunting bats
We establish an agent-based model to study the impact of prey behavior on the hunting success of predators. The predators and prey are modeled as self-propelled particles moving in a three-dimensional domain and subject to specific sensing abilities and behavioral rules inspired by bat hunting. The predators randomly search for prey. The prey either align velocity directions with peers, defined as “interacting” prey, or swarm “independently” of peer presence; both types of prey are subject to additive noise. In a simulation study, we find that interacting prey using low noise have the maximum predation avoidance because they form localized large groups, while they suffer high predation as noise increases due to the formation of broadly dispersed small groups. Independent prey, which are likely to be uniformly distributed in the domain, have higher predation risk under a low noise regime as they traverse larger spatial extents. These effects are enhanced in large prey populations, which exhibit more ordered collective behavior or more uniform spatial distribution as they are interacting or independent, respectively.