Blinded sample size review for McNemar's test based on primary and surrogate endpoints
本文针对配对二元主要和短期替代终点，开发了McNemar检验的盲样本大小重估策略，以保持I类错误率同时达到预定目标功效。
0 citationsRead paper
Selected work
Representative Papers
Sterilizable Scene Graph Generation for Operating Rooms
This study addresses the challenges of parameter redundancy, deployment difficulties, and privacy risks in operating room scene graph generation by proposing the SG-NCA framework. This approach introduces a novel paradigm integrating Neural Cellular Automata (NCA) for scene graph generation and structured representation learning, combined with multi-class segmentation and a lightweight relation predictor. Experimental results demonstrate that SG-NCA achieves performance comparable to mainstream baselines while reducing model parameters by 55 times. Consequently, it enables successful deployment on fanless edge devices, effectively satisfying sterile environment requirements and ensuring data privacy. These findings establish SG-NCA as a viable solution for lightweight medical AI applications, offering a new pathway for secure and efficient intraoperative analysis without compromising accuracy or safety standards in clinical settings.
0 citationsRead paper
From Asymptotic to Finite-Sample Minimax Robust Hypothesis Testing
This work addresses the finite-sample minimax robust hypothesis testing problem under distributional uncertainty by rigorously bridging finite-sample and asymptotic optimal solutions through asymptotic theory, thereby avoiding conventional heuristic constructions. The authors model uncertainty using total variation distance and band models, and derive explicit parametric forms for the least favorable distributions and the robust likelihood ratio function. Theoretical analysis establishes that, under both uncertainty models, the finite-sample minimax robust test coincides with its asymptotic counterpart, extending existing results to asymmetric robust parameter settings and providing a systematic unification of prior approaches. Numerical simulations corroborate the theoretical guarantees and demonstrate the practical efficacy of the proposed test.
0 citationsRead paper
Asymptotically Minimax Robust Likelihood Ratio Test
This study addresses hypothesis testing under distributional uncertainty by developing asymptotically minimax robust tests within both Bayesian and Neyman–Pearson frameworks. The authors characterize uncertainty sets using Kullback–Leibler divergence, α-divergence, and their symmetric variants, and—by leveraging Sion’s minimax theorem and Karush–Kuhn–Tucker conditions—derive, for the first time, closed-form solutions for robust likelihood ratio tests under various divergence constraints. They rigorously establish the existence and uniqueness of these solutions. Theoretical analysis further reveals that the Dabak method fails to achieve asymptotic minimax robustness, thereby correcting and extending existing theory. Numerical experiments confirm the effectiveness and superiority of the proposed approach.
0 citationsRead paper
Effect of Reporting Mode and Clinical Experience on Radiologists' Gaze and Image Analysis Behavior in Chest Radiography
This study investigates how free-text reporting, structured reporting, and AI-assisted structured reporting affect radiologists’ interpretation of chest X-rays. Using eye-tracking and a custom阅片 platform, we quantified diagnostic accuracy, reporting efficiency, visual behavior (e.g., saccade count, dwell time on report regions), and user experience via generalized linear mixed models with Bonferroni correction, while examining the moderating effect of radiologist experience. Results show that AI-assisted structured reporting significantly improved diagnostic consistency (Cohen’s κ = 0.71), reduced reporting time (25 ± 9 seconds), lowered cognitive load (23% fewer saccades; 31% shorter dwell time in report regions), and received the highest user preference. Novice and experienced radiologists exhibited distinct attentional allocation patterns. This is the first study to elucidate the cognitive mechanisms underlying intelligent reporting systems using multimodal behavioral data, providing empirical evidence and theoretical grounding for human–AI collaborative reporting design in clinical practice.
0 citationsRead paper
Recent publications
Latest Papers
Blinded sample size review for McNemar's test based on primary and surrogate endpoints
本文针对配对二元主要和短期替代终点，开发了McNemar检验的盲样本大小重估策略，以保持I类错误率同时达到预定目标功效。
0 citationsRead paper
Sterilizable Scene Graph Generation for Operating Rooms
This study addresses the challenges of parameter redundancy, deployment difficulties, and privacy risks in operating room scene graph generation by proposing the SG-NCA framework. This approach introduces a novel paradigm integrating Neural Cellular Automata (NCA) for scene graph generation and structured representation learning, combined with multi-class segmentation and a lightweight relation predictor. Experimental results demonstrate that SG-NCA achieves performance comparable to mainstream baselines while reducing model parameters by 55 times. Consequently, it enables successful deployment on fanless edge devices, effectively satisfying sterile environment requirements and ensuring data privacy. These findings establish SG-NCA as a viable solution for lightweight medical AI applications, offering a new pathway for secure and efficient intraoperative analysis without compromising accuracy or safety standards in clinical settings.
0 citationsRead paper
From Asymptotic to Finite-Sample Minimax Robust Hypothesis Testing
This work addresses the finite-sample minimax robust hypothesis testing problem under distributional uncertainty by rigorously bridging finite-sample and asymptotic optimal solutions through asymptotic theory, thereby avoiding conventional heuristic constructions. The authors model uncertainty using total variation distance and band models, and derive explicit parametric forms for the least favorable distributions and the robust likelihood ratio function. Theoretical analysis establishes that, under both uncertainty models, the finite-sample minimax robust test coincides with its asymptotic counterpart, extending existing results to asymmetric robust parameter settings and providing a systematic unification of prior approaches. Numerical simulations corroborate the theoretical guarantees and demonstrate the practical efficacy of the proposed test.
0 citationsRead paper
Asymptotically Minimax Robust Likelihood Ratio Test
This study addresses hypothesis testing under distributional uncertainty by developing asymptotically minimax robust tests within both Bayesian and Neyman–Pearson frameworks. The authors characterize uncertainty sets using Kullback–Leibler divergence, α-divergence, and their symmetric variants, and—by leveraging Sion’s minimax theorem and Karush–Kuhn–Tucker conditions—derive, for the first time, closed-form solutions for robust likelihood ratio tests under various divergence constraints. They rigorously establish the existence and uniqueness of these solutions. Theoretical analysis further reveals that the Dabak method fails to achieve asymptotic minimax robustness, thereby correcting and extending existing theory. Numerical experiments confirm the effectiveness and superiority of the proposed approach.
0 citationsRead paper
Effect of Reporting Mode and Clinical Experience on Radiologists' Gaze and Image Analysis Behavior in Chest Radiography
This study investigates how free-text reporting, structured reporting, and AI-assisted structured reporting affect radiologists’ interpretation of chest X-rays. Using eye-tracking and a custom阅片 platform, we quantified diagnostic accuracy, reporting efficiency, visual behavior (e.g., saccade count, dwell time on report regions), and user experience via generalized linear mixed models with Bonferroni correction, while examining the moderating effect of radiologist experience. Results show that AI-assisted structured reporting significantly improved diagnostic consistency (Cohen’s κ = 0.71), reduced reporting time (25 ± 9 seconds), lowered cognitive load (23% fewer saccades; 31% shorter dwell time in report regions), and received the highest user preference. Novice and experienced radiologists exhibited distinct attentional allocation patterns. This is the first study to elucidate the cognitive mechanisms underlying intelligent reporting systems using multimodal behavioral data, providing empirical evidence and theoretical grounding for human–AI collaborative reporting design in clinical practice.
0 citationsRead paper