Listening for Airway Stenosis: A Foundation Model-Based Method for Rapid and Accessible Detection
研究利用声学AI和患者语音记录快速检测气道狭窄，通过声学基础模型提取相关声学特征，实现高精度筛查。
0 citationsRead paper
Selected work
Representative Papers
Auditable CT Phenotyping Through Report-derived Radiological Observations
研究使用基于报告的放射学观察构建的可审计CT表型（ACT）方法，解决了医学图像基础模型在预测临床表型时依赖疾病特异性发现还是捷径的问题。
0 citationsRead paper
FRAME: separating sampling variation from representational cause in medical imaging fairness
本文提出FRAME框架，通过两步法区分医学影像公平性中的采样变异和表示原因，从而更准确地评估和解决公平性问题。
0 citationsRead paper
Privacy-Preserving Detection of Rare Disease-Associated Cell Subsets via Secure Multi-Party Computation
本文提出一种安全多方计算框架，用于在保护隐私的前提下训练和应用CellCnn模型以识别罕见疾病相关的细胞亚群。
0 citationsRead paper
Bayesian uncertainty estimation improves clinical decision making in medical AI agents
This study addresses the critical limitation of medical AI systems—their inability to provide reliable confidence assessments for ambiguous or atypical cases, which hinders clinical deployment. The authors propose integrating Monte Carlo Dropout into a multi-task chest X-ray classifier to estimate epistemic uncertainty and demonstrate, for the first time, that this uncertainty signal effectively enhances clinical decision support. A key innovation lies in incorporating uncertainty as a binary risk flag rather than raw scores, substantially improving practical utility. Experimental results show that this approach increases the AUROC for error detection from 0.74 to 0.77 and reduces the high-confidence misdiagnosis rate from 8.5% to 2.7% in controlled testing, highlighting its potential to improve safety and reliability in real-world clinical settings.
0 citationsRead paper
Recent publications
Latest Papers
Listening for Airway Stenosis: A Foundation Model-Based Method for Rapid and Accessible Detection
研究利用声学AI和患者语音记录快速检测气道狭窄，通过声学基础模型提取相关声学特征，实现高精度筛查。
0 citationsRead paper
Auditable CT Phenotyping Through Report-derived Radiological Observations
研究使用基于报告的放射学观察构建的可审计CT表型（ACT）方法，解决了医学图像基础模型在预测临床表型时依赖疾病特异性发现还是捷径的问题。
0 citationsRead paper
FRAME: separating sampling variation from representational cause in medical imaging fairness
本文提出FRAME框架，通过两步法区分医学影像公平性中的采样变异和表示原因，从而更准确地评估和解决公平性问题。
0 citationsRead paper
Privacy-Preserving Detection of Rare Disease-Associated Cell Subsets via Secure Multi-Party Computation
本文提出一种安全多方计算框架，用于在保护隐私的前提下训练和应用CellCnn模型以识别罕见疾病相关的细胞亚群。
0 citationsRead paper
Bayesian uncertainty estimation improves clinical decision making in medical AI agents
This study addresses the critical limitation of medical AI systems—their inability to provide reliable confidence assessments for ambiguous or atypical cases, which hinders clinical deployment. The authors propose integrating Monte Carlo Dropout into a multi-task chest X-ray classifier to estimate epistemic uncertainty and demonstrate, for the first time, that this uncertainty signal effectively enhances clinical decision support. A key innovation lies in incorporating uncertainty as a binary risk flag rather than raw scores, substantially improving practical utility. Experimental results show that this approach increases the AUROC for error detection from 0.74 to 0.77 and reduces the high-confidence misdiagnosis rate from 8.5% to 2.7% in controlled testing, highlighting its potential to improve safety and reliability in real-world clinical settings.
0 citationsRead paper