Data-driven individualized decision making has recently received increasing research interest. However, most existing methods rely on the assumption of no unmeasured confounding, which cannot be ensured in practice especially in observational studies. Motivated by the recently proposed proximal causal inference, we develop several proximal learning methods to estimate optimal individualized treatment regimes (ITRs) in the presence of unmeasured confounding. Explicitly, in terms of two types of proxy variables, we are able to establish several identification results for different classes of ITRs respectively, exhibiting the trade-off between the risk of making untestable assumptions and the potential improvement of the value function in decision making. Based on these identification results, we propose several classification-based approaches to finding a variety of restricted in-class optimal ITRs and establish their theoretical properties. The appealing numerical performance of our proposed methods is demonstrated via extensive simulation experiments and a real data application.
Neonatal Endotracheal Intubation (ETI) is a critical resuscitation skill that requires tremendous practice of trainees before clinical exposure. However, current manikin-based training regimen is ineffective in providing satisfactory real-time procedural guidance for accurate assessment due to the lack of see-through visualization within the manikin. The training efficiency is further reduced by the limited availability of expert instructors, which inevitably results in a long learning curve for trainees. To this end, we propose an intelligent Augmented Reality (AR) training framework that provides trainees with a complete visualization of the ETI procedure for real-time guidance and assessment. Specifically, the proposed framework is capable of capturing the motions of the laryngoscope and the manikin and offer 3D see-through visualization rendered to the head-mounted display (HMD). Furthermore, an attention-based Convolutional Neural Network (CNN) model is developed to automatically assess the ETI performance from the captured motions as well as identify regions of motions that significantly contribute to the performance evaluation. Lastly, augmented user-friendly feedback is delivered with interpretable results with the ETI scoring rubric through the color-coded motion trajectory that classifies highlighted regions that need more practice. The classification accuracy of our machine learning model is 84.6%.
We present results from a computational analysis developed to integrate transcriptome and metabolomic data in order to explore the heat stress response in the liver of the modern broiler chicken. Heat stress is a significant cause of productivity loss in the poultry industry, both in terms of increased livestock morbidity and its negative influence on average feed efficiency. This study focuses on the liver because it is an important regulator of metabolism, controlling many of the physiological processes impacted by prolonged heat stress. Using statistical learning methods, we identify genes and metabolites that may regulate the heat stress response in the liver and adaptations required to acclimate to prolonged heat stress. We describe how disparate systems such as sugar, lipid and amino acid metabolism, are coordinated during the heat stress response. Our findings provide more detailed context for genomic studies and generates hypotheses about dietary interventions that can mitigate the negative influence of heat stress on the poultry industry.
15 April 2024 in Journal of Computational and Graphical Statistics
Body composition can be assessed in many different ways. High-end medical equipment, such as Dual-energy X-ray Absorptiometry (DXA), Computed Tomography (CT), and Magnetic Resonance Imaging (MRI) offers high-fidelity pixel/voxel-level assessment, but is prohibitive in cost. In the case of DXA and CT, the approach exposes users to ionizing radiation. Whole-body air displacement plethysmography (BOD POD) can accurately estimate body density, but the assessment is limited to the whole-body fat percentage. Optical three-dimensional (3D) scan and reconstruction techniques, such as using depth cameras, have brought new opportunities for improving body composition assessment by intelligently analyzing body shape features. In this paper, we present a novel supervised inference model to predict pixel-level body composition and percentage of body fat using 3D geometry features and body density. First, we use body density to model a fat distribution base prediction. Then, we use a Bayesian network to infer the probability of the base prediction bias with 3D geometry features. Finally, we correct the bias using non-parametric regression. We use DXA assessment as the ground truth in model training and validation. We compare our method, in terms of pixel-level body composition assessment, with the current state-of-the-art prediction models. Our method outperforms those prediction models by 52.69% on average. We also compare our method, in terms of whole-body fat percentage assessment, with the medical-level equipment-BOD POD. Our method outperforms the BOD POD by 23.28%.
1 November 2021 in Computational Statistics & Data Analysis