To develop an intelligent diagnostic model for osteoporosis screening based on low-dose chest computed tomography (LDCT). The model incorporates automatic deep-learning thoracic vertebrae of cancellous bone (TVCB) segmentation model and radiomics analysis. A total of 442 participants who underwent both LDCT and quantitative computed tomography (QCT) examinations were enrolled and were randomly allocated to the training, internal testing, and external testing cohorts. The TVCB automatic segmentation model was trained using VB-Net. The accuracy of the segmentation was evaluated using the Dice coefficient. Predictive models for assessing bone mineral density (BMD) were constructed utilizing radiomics analysis based on automatic segmentation (ASeg model) and manual segmentation (MSeg model), respectively. The BMD predictive model based on ASeg and MSeg included the identification of normal and abnormal BMD (first-level model), and osteopenia and osteoporosis (second-level model). The diagnostic performance of the radiomics models were evaluated using the area under the curve (AUC), sensitivity and specificity. The Dice coefficients of the TVCB segmentation model in the internal and external testing cohorts were found to be 0.988 ± 0.014 and 0.939 ± 0.034, respectively. In the first-level model, the AUC of the ASeg model exhibited comparable performance to that of the MSeg model for both the internal (0.985 vs. 0.946, P = 0.080) and external (0.965 vs. 0.955, P = 0.724) testing cohorts. Similarly, in the second-level model, the AUC of the ASeg model was found to be comparable to that of the MSeg model for both the internal (0.933 vs. 0.920, P = 0.794) and external (0.907 vs. 0.892, P = 0.805) testing cohorts. A fully automated pipeline for TVCB segmentation and BMD assessment with radiomics analysis can be used for opportunistic BMD screening in chest LDCT.

Yinhao Bai, Zhixin Han, Yuhua Zhao, Hang Gao, Zhuowei Zhang, Xunzhi Wang, Mengting Hu. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024.

Traffic accidents caused by driver distractions have seriously endangered public safety, with driver distractions typically stemming from behaviors beyond safe driving. Recently, vision-based driver behavior recognition has attracted much attention, achieving great success with deep learning-based schemes. However, the generalization ability of these models in real-world scenarios remains unsatisfactory. In this paper, we conduct an in-depth investigation into the underlying causes of this unsatisfactory generalization and conclude that the behavior features extracted by convolutional neural networks are intertwined with driver identity features. Based on this discovery, we propose a feature decomposition (FD) framework to disentangle these two types of features. The separated behavior features, referred to as driver-irrelevant behavior features, are subsequently leveraged for behavior recognition. Moreover, we introduce a co-training strategy to optimize the FD framework. This strategy enables behavior features and identity features to provide mutual auxiliary signals and encourages each other to drop the information that do not belong to them, so that the learned behavior features can be driver-irrelevant. Rigorous experiments are conducted on two widely-studied datasets, yielding results that demonstrate the superior performance of our framework and its improved generalization capabilities. Importantly, our framework’s fast inference capabilities make it highly suitable for real-world scenarios. Codes are released at https://github.com/gaohangcodes/ LearningDriverIrrelevantFeatures4DBR.

Rui Ying, Mengting Hu, Jianfeng Wu, Yalan Xie, Xiaoyi Liu, Zhunheng Wang, Ming Jiang, Hang Gao, Linlin Zhang, Renhong Cheng. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.