Shi Gu
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
Research Subjects
Shi Gu has a total of 173 co-authors. Here are the top co-authors by number of shared publications:
- Danielle S. Bassett
Yale University
9 shared publications Yifan Wu
Department of Bioengineering, University of Pennsylvania, Philadelphia, USA
9 shared publications- Su Lui
Huaxi MR Research Center (HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu, China
7 shared publications Huzheng Yang
School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China; Department of Computer and Information Science, University of Pennsylvania, Philadelphia, PA 19104, USA
6 shared publicationsTianwei Zhang
School of Computer and Engineering, University of Electronic Science and Technology of China, UESTC, China
6 shared publicationsJames C. Gee
University of Pennsylvania, Philadelphia, USA
5 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12949-12958
Event-Guided Super-Resolution for Novel View Synthesis in Dynamic Scenes
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12949-12958
Open AccessEvent-Guided Super-Resolution for Novel View Synthesis in Dynamic Scenes
- 28 September 2026
- Published by Springer Nature in Nature Machine Intelligence
- Vol. 8 (10) , 1548-1559
Task-structured modularity emerges in artificial networks and aligns with brain architecture
- 28 September 2026
- Published by Springer Nature in Nature Machine Intelligence
- Vol. 8 (10) , 1548-1559
Task-structured modularity emerges in artificial networks and aligns with brain architecture
- Open Access
- 23 April 2026
- Published by Springer Nature in BMC Medicine
- Vol. 24 (1)
Multiparameter concept-based interpretable model for early breast cancer diagnosis and structured reporting: a multi-center, multi-reader, radiologist-in-the-loop study
Accurately differentiating early-stage breast cancer from benign lesions on MRI is essential to reduce unnecessary biopsies. However, the limited interpret
- 23 April 2026
- Published by Springer Nature in BMC Medicine
- Vol. 24 (1)
Open AccessMultiparameter concept-based interpretable model for early breast cancer diagnosis and structured reporting: a multi-center, multi-reader, radiologist-in-the-loop study
Accurately differentiating early-stage breast cancer from benign lesions on MRI is essential to reduce unnecessary biopsies. However, the limited interpret
- Open Access
- 18 April 2026
- Published by Springer Nature in Scientific Data
- Vol. 13 (1)
A large-scale fMRI dataset for vision-language semantic association
Understanding the neural coding and association of visual and language information benefits from the development of deep learning models and the collection of massive datasets with extensive sampling of brain activity. Large-scale functional magnetic resonance imaging (fMRI) datasets with naturalistic stimuli provide more ecologically relevant experimental conditions and promote more reproducible research into the neural basis of sensory perception. Here, unlike most previous datasets restricted to isolated modalities, we present the Caption Scene Dataset (CSD), a large-scale fMRI dataset for vision-language semantic association, in which neural responses to 4,400 pairs of Chinese captions and naturalistic scenes were acquired from eight healthy participants. The participants were instructed to determine whether the semantics in the caption and the image are consistent. To illustrate the utility of the CSD dataset, we demonstrated that deep neural encoding models effectively predicted neural responses to both caption and image stimuli across different cortical regions. This dataset provides a platform for the investigation of the neural basis of semantic association across vision and language, facilitating cross-disciplinary advances between vision neuroscience and artificial intelligence.
- 18 April 2026
- Published by Springer Nature in Scientific Data
- Vol. 13 (1)
Open AccessA large-scale fMRI dataset for vision-language semantic association
Understanding the neural coding and association of visual and language information benefits from the development of deep learning models and the collection of massive datasets with extensive sampling of brain activity. Large-scale functional magnetic resonance imaging (fMRI) datasets with naturalistic stimuli provide more ecologically relevant experimental conditions and promote more reproducible research into the neural basis of sensory perception. Here, unlike most previous datasets restricted to isolated modalities, we present the Caption Scene Dataset (CSD), a large-scale fMRI dataset for vision-language semantic association, in which neural responses to 4,400 pairs of Chinese captions and naturalistic scenes were acquired from eight healthy participants. The participants were instructed to determine whether the semantics in the caption and the image are consistent. To illustrate the utility of the CSD dataset, we demonstrated that deep neural encoding models effectively predicted neural responses to both caption and image stimuli across different cortical regions. This dataset provides a platform for the investigation of the neural basis of semantic association across vision and language, facilitating cross-disciplinary advances between vision neuroscience and artificial intelligence.
- Open Access
- 31 July 2025
- Published by Springer Nature in Research Square
Task-structured Modularity Emerges in Artificial Networks and Aligns with Brain Architecture
Understanding how neural systems develop modular organization is fundamental to both neuroscience and artificial intelligence. Although modular architectures can improve adaptability and cognitive performance, the processes leading to their emergence are poorly understood. Here we demonstrate that multitask and incremental learning enhance modularity in recurrent neural networks (RNNs) compared to single-task learning, revealing how functional demands influence the structural organization of neural networks. We trained RNNs on cognitive tasks under three distinct learning paradigms: single-task, simultaneous multitask, and incremental multitask learning. Our results suggest that networks adopting multitask learning show an enhanced degree of modularity compared to single-task training, especially when the task load exceeds the network's representation capacity. Those trained with incremental multitask learning, in particular, develop the highest degree of modularity, maintaining superior performance even when connections are selectively pruned. Furthermore, these task-induced networks exhibit structural properties more closely resembling those of biological brain networks than those based solely on spatial constraints, particularly in clustering coefficients and edge length distributions. Collectively, these findings suggest that modular brain architecture emerges not only from physical constraints but as an adaptive response to the sequential introduction of complex cognitive tasks, providing a causal explanation for how functional demands shape network topology. Our findings also illustrate how the developmental and evolutionary context of the brain, wherein multiple tasks are learned, can inform the design of artificial systems well-suited for similar environments.
- 31 July 2025
- Published by Springer Nature in Research Square
Open AccessTask-structured Modularity Emerges in Artificial Networks and Aligns with Brain Architecture
Understanding how neural systems develop modular organization is fundamental to both neuroscience and artificial intelligence. Although modular architectures can improve adaptability and cognitive performance, the processes leading to their emergence are poorly understood. Here we demonstrate that multitask and incremental learning enhance modularity in recurrent neural networks (RNNs) compared to single-task learning, revealing how functional demands influence the structural organization of neural networks. We trained RNNs on cognitive tasks under three distinct learning paradigms: single-task, simultaneous multitask, and incremental multitask learning. Our results suggest that networks adopting multitask learning show an enhanced degree of modularity compared to single-task training, especially when the task load exceeds the network's representation capacity. Those trained with incremental multitask learning, in particular, develop the highest degree of modularity, maintaining superior performance even when connections are selectively pruned. Furthermore, these task-induced networks exhibit structural properties more closely resembling those of biological brain networks than those based solely on spatial constraints, particularly in clustering coefficients and edge length distributions. Collectively, these findings suggest that modular brain architecture emerges not only from physical constraints but as an adaptive response to the sequential introduction of complex cognitive tasks, providing a causal explanation for how functional demands shape network topology. Our findings also illustrate how the developmental and evolutionary context of the brain, wherein multiple tasks are learned, can inform the design of artificial systems well-suited for similar environments.
- Open Access
- 12 May 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1
Author Correction: A concept-based interpretable model for the diagnosis of choroid neoplasias using multimodal data
- 12 May 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1
Open AccessAuthor Correction: A concept-based interpretable model for the diagnosis of choroid neoplasias using multimodal data
- Top Cited PapersOpen Access
- 13 April 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-14
A concept-based interpretable model for the diagnosis of choroid neoplasias using multimodal data
Diagnosing rare diseases remains a critical challenge in clinical practice, often requiring specialist expertise. Despite the promising potential of machine learning, the scarcity of data on rare diseases and the need for interpretable, reliable artificial intelligence (AI) models complicates development. This study introduces a multimodal concept-based interpretable model tailored to distinguish uveal melanoma (0.4-0.6 per million in Asians) from hemangioma and metastatic carcinoma following the clinical practice. We collected a comprehensive dataset on Asians to date on choroid neoplasm imaging with radiological reports, encompassing over 750 patients from 2013 to 2019. Our model integrates domain expert insights from radiological reports and differentiates between three types of choroidal tumors, achieving an F1 score of 0.91. This performance not only matches senior ophthalmologists but also improves the diagnostic accuracy of less experienced clinicians by 42%. The results underscore the potential of interpretable AI to enhance rare disease diagnosis and pave the way for future advancements in medical AI.
- 13 April 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-14
Top Cited PapersOpen AccessA concept-based interpretable model for the diagnosis of choroid neoplasias using multimodal data
Diagnosing rare diseases remains a critical challenge in clinical practice, often requiring specialist expertise. Despite the promising potential of machine learning, the scarcity of data on rare diseases and the need for interpretable, reliable artificial intelligence (AI) models complicates development. This study introduces a multimodal concept-based interpretable model tailored to distinguish uveal melanoma (0.4-0.6 per million in Asians) from hemangioma and metastatic carcinoma following the clinical practice. We collected a comprehensive dataset on Asians to date on choroid neoplasm imaging with radiological reports, encompassing over 750 patients from 2013 to 2019. Our model integrates domain expert insights from radiological reports and differentiates between three types of choroidal tumors, achieving an F1 score of 0.91. This performance not only matches senior ophthalmologists but also improves the diagnostic accuracy of less experienced clinicians by 42%. The results underscore the potential of interpretable AI to enhance rare disease diagnosis and pave the way for future advancements in medical AI.
- 23 March 2025
- Published by Elsevier in Medical Image Analysis
- Vol. 102, 103543
Enhancing source-free domain adaptation in Medical Image Segmentation via regulated model self-training
- 23 March 2025
- Published by Elsevier in Medical Image Analysis
- Vol. 102, 103543
Enhancing source-free domain adaptation in Medical Image Segmentation via regulated model self-training
- 21 October 2024
- Published by Elsevier in Journal of Affective Disorders
- Vol. 370, 268-276
Transdiagnostic white matter controllability deficits across patients with affective and anxiety spectrum disorders
- 21 October 2024
- Published by Elsevier in Journal of Affective Disorders
- Vol. 370, 268-276
Transdiagnostic white matter controllability deficits across patients with affective and anxiety spectrum disorders
- Open Access
- 26 July 2024
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 10 (30) , eadm8430
Emergence and reconfiguration of modular structure for artificial neural networks during continual familiarity detection
Advances in artificial intelligence enable neural networks to learn a wide variety of tasks, yet our understanding of the learning dynamics of these networks remains limited. Here, we study the temporal dynamics during learning of Hebbian feedforward neural networks in tasks of continual familiarity detection. Drawing inspiration from network neuroscience, we examine the network’s dynamic reconfiguration, focusing on how network modules evolve throughout learning. Through a comprehensive assessment involving metrics like network accuracy, modular flexibility, and distribution entropy across diverse learning modes, our approach reveals various previously unknown patterns of network reconfiguration. We find that the emergence of network modularity is a salient predictor of performance and that modularization strengthens with increasing flexibility throughout learning. These insights not only elucidate the nuanced interplay of network modularity, accuracy, and learning dynamics but also bridge our understanding of learning in artificial and biological agents.
- 26 July 2024
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 10 (30) , eadm8430
Open AccessEmergence and reconfiguration of modular structure for artificial neural networks during continual familiarity detection
Advances in artificial intelligence enable neural networks to learn a wide variety of tasks, yet our understanding of the learning dynamics of these networks remains limited. Here, we study the temporal dynamics during learning of Hebbian feedforward neural networks in tasks of continual familiarity detection. Drawing inspiration from network neuroscience, we examine the network’s dynamic reconfiguration, focusing on how network modules evolve throughout learning. Through a comprehensive assessment involving metrics like network accuracy, modular flexibility, and distribution entropy across diverse learning modes, our approach reveals various previously unknown patterns of network reconfiguration. We find that the emergence of network modularity is a salient predictor of performance and that modularization strengthens with increasing flexibility throughout learning. These insights not only elucidate the nuanced interplay of network modularity, accuracy, and learning dynamics but also bridge our understanding of learning in artificial and biological agents.