Large-Scale Pretraining for Improving Deep Learning-Based Geometric Distortion Correction of Diffusion-Weighted Imaging
研究利用大规模预训练策略改进基于深度学习的扩散加权成像几何失真校正，通过自监督和生成式预训练模型提升校正效果。
0 citationsRead paper
Selected work
Representative Papers
FoldKit: A Python library for efficient storage and retrieval of co-folding predictions
为解决AlphaFold 3共折叠实验产生的大量数据存储问题，开发了Python库FoldKit，通过高效压缩和结构化处理这些数据，并提供便捷的程序访问接口。
0 citationsRead paper
A Source-Grounded Framework for Constructing and Evaluating Progressive Multimodal Diagnostic Dialogues from Clinical Case Reports
本文提出一种基于源的框架，用于从病例报告中构建和评估渐进式多模态诊断对话，解决现有方法无法有效整合逐步临床证据的问题。
0 citationsRead paper
LLM assisted writing deserves empirical evaluation
研究通过分析69,209篇健康信息学论文，评估了LLM辅助写作的影响，提倡基于学术质量和责任而非工具使用来评价论文。
0 citationsRead paper
GeoPose: Patient-agnostic CTA-to-DSA registration through projection-space calibration
This study addresses the challenges of initialization dependency and patient-specific training in intraoperative DSA-CTA registration by proposing GeoPose, a novel framework employing population-level training and projection-space calibration to align poses within the native coordinate system. By integrating residual networks with lightweight optimization, GeoPose achieves cross-individual generalization without patient adaptation and enables direct registration in native frames. Experimental results demonstrate that without optimization, the mean centerline distance (mPCD) is 5.8 mm (0.15 s); after 25 iterations, mPCD decreases to 4.6 mm and clDice increases to 0.58, significantly outperforming baseline methods. These improvements effectively support downstream biplane vascular reconstruction, offering a robust solution for real-time clinical applications.
0 citationsRead paper
Recent publications
Latest Papers
Large-Scale Pretraining for Improving Deep Learning-Based Geometric Distortion Correction of Diffusion-Weighted Imaging
研究利用大规模预训练策略改进基于深度学习的扩散加权成像几何失真校正，通过自监督和生成式预训练模型提升校正效果。
0 citationsRead paper
FoldKit: A Python library for efficient storage and retrieval of co-folding predictions
为解决AlphaFold 3共折叠实验产生的大量数据存储问题，开发了Python库FoldKit，通过高效压缩和结构化处理这些数据，并提供便捷的程序访问接口。
0 citationsRead paper
A Source-Grounded Framework for Constructing and Evaluating Progressive Multimodal Diagnostic Dialogues from Clinical Case Reports
本文提出一种基于源的框架，用于从病例报告中构建和评估渐进式多模态诊断对话，解决现有方法无法有效整合逐步临床证据的问题。
0 citationsRead paper
LLM assisted writing deserves empirical evaluation
研究通过分析69,209篇健康信息学论文，评估了LLM辅助写作的影响，提倡基于学术质量和责任而非工具使用来评价论文。
0 citationsRead paper
GeoPose: Patient-agnostic CTA-to-DSA registration through projection-space calibration
This study addresses the challenges of initialization dependency and patient-specific training in intraoperative DSA-CTA registration by proposing GeoPose, a novel framework employing population-level training and projection-space calibration to align poses within the native coordinate system. By integrating residual networks with lightweight optimization, GeoPose achieves cross-individual generalization without patient adaptation and enables direct registration in native frames. Experimental results demonstrate that without optimization, the mean centerline distance (mPCD) is 5.8 mm (0.15 s); after 25 iterations, mPCD decreases to 4.6 mm and clDice increases to 0.58, significantly outperforming baseline methods. These improvements effectively support downstream biplane vascular reconstruction, offering a robust solution for real-time clinical applications.
0 citationsRead paper