Recent advances in sensing and imaging technologies have enabled the acquisition of high-dimensional spatiotemporal data from complex geometric domains. However, predictive modeling of such systems remains challenging due to irregular spatial manifolds, coupled multi-output dynamics, limited observations, and the need for reliable uncertainty quantification. This paper presents a physics-augmented, geometry-aware, multi-task Gaussian Process (P-G-MGP) framework for spatiotemporal modeling. We develop a geometry-aware multi-task GP (G-MGP) to capture spatiotemporal structures and inter-task dependencies. To enhance model fidelity and robustness, we incorporate governing physical laws through a physics-based regularization scheme, thereby constraining predictions to be consistent with governing principles. Furthermore, our framework provides closed-form estimates of posterior variance, enabling calibrated uncertainty quantification for downstream decision-making. We validate P-G-MGP on 3D cardiac electrophysiological modeling, demonstrating superior predictive performance over existing methods by effectively incorporating geometric priors, multi-task interactions, and domain-specific physical constraints.
Federated Learning (FL) is an emerging computing paradigm to collaboratively train Machine Learning (ML) models across multi-source data while preserving privacy. The major challenge of “meaningful” implementation of FL for any ML model is how to guarantee that the federated ML model can achieve comparable performance compared to the global model trained using the combined data. Moreover, there are very limited studies on FL of functional regression models that analyze functional data, a commonly encountered type of data in many fields. This study develops the first-of-its-kind federated Gradient Boosting algorithm with the Least Squares Approximation (fed-GB-LSA) for efficient, privacy-preserving federated learning of the function-on-function regression with several distinct merits: (1) The GB-based algorithm allows the sparse selection of multivariate functional and non-functional features in the function-on-function regression prediction, which is not straightforward in the functional regression; (2) The parameter estimation by the GB algorithm results in separate sub-optimization problems with explicitly analytical solutions for each of the features, providing an efficient estimation algorithm for the function-on-function regression; (3) The LSA-enabled fed-GB provides a “one-shot” approach for FL that is communicationally and statistically efficient, providing theoretical guarantees to the federated model’s performance without data sharing across local servers. The proposed fed-GB-LSA is tested in extensive simulation studies by considering real-world challenges such as device heterogeneity and applied in a real-world dataset for privacy-preserving telemonitoring of Obstructive Sleep Apnea (OSA).
The era of big data has made vast amounts of clinical data readily available, particularly in the form of electronic health records (EHRs), which provides unprecedented opportunities for developing data-driven diagnostic tools to enhance clinical decision making. However, data-driven modeling of EHRs faces challenges such as irregularly spaced time series, issues of incompleteness, and data imbalance. Realizing the full data potential of EHRs hinges on the development of advanced analytical models. In this paper, we propose a novel Missingness-aware mUlti-branching Self-Attention Encoder (MUSE-Net) to cope with the challenges in modeling longitudinal EHRs for data-driven disease prediction. The proposed MUSE-Net is composed by four novel modules including: 1) a multi-task Gaussian process (MGP) with missing value masks for data imputation; 2) a multi-branching architecture to address the data imbalance problem; 3) a time-aware self-attention encoder to account for the irregularly spaced time interval in longitudinal EHRs; 4) interpretable multi-head attention mechanism that provides insights into the importance of different time points in disease prediction, allowing clinicians to trace model decisions. We evaluate the proposed MUSE-Net using both synthetic and real-world datasets. Experimental results show that our MUSE-Net outperforms existing methods that are widely used to investigate longitudinal signals.
Sleep apnea, a prevalent sleep-related breathing disorder, often remains undiagnosed and untreated in a large patient population due to the need of extensive manual annotations on various physiological signals for clinical diagnosis. Despite the surge of interest in applying machine learning to automate apnea detection, the effectiveness of existing techniques highly relies on strongly supervised learning that requires massive finely labeled training data for sufficiently short time intervals - a requirement often unmet due to the prohibitively high cost of manual labeling in clinical practice. In this article, we incorporate clinical knowledge to establish a weakly supervised deep learning framework for automatically estimating the latent fine-grained apnea severity when only coarse-grained labels indicating apnea presence are available in the training data. Specifically, a novel knowledge-enhanced dual-granularity consistency loss, which simultaneously considers the consistency between coarse- and fine-granularity and the integration of clinical knowledge on apnea diagnosis, is designed to boost the model's learning of apnea severity at the fine granularity. A mathematical encoding of clinical knowledge is proposed to calibrate fine-grained estimation accuracy through ordinal alignment functions, which quantitatively relates the severity of apnea to the prominence of key diagnosis-informed physiological symptoms. The proposed method is able to accurately estimate fine-grained apnea severity in real time with significantly reduced labeling costs, extending the reach of sleep apnea diagnostics to larger population both in lab and at home. An experiment is conducted to demonstrate the superior estimation performance of the proposed method for monitoring apnea severity at high temporal resolution.
Cone beam computed tomography (CBCT) is a widely-used imaging modality in dental healthcare. It is an important task to segment each 3D CBCT image, which involves labeling lesions, bones, teeth, and restorative materials on a voxel-by-voxel basis, as it aids in lesion detection, diagnosis, and treatment planning. The current clinical practice relies on manual segmentation, which is labor-intensive and demands considerable expertise. Leveraging Artificial Intelligence (AI) to fully automate the segmentation process could tremendously improve the quality and efficiency of dental healthcare. The main hurdle in this advancement is reducing AI's reliance on a large quantity of manually labeled images to train robust, accurate, and generalizable algorithms. To tackle this challenge, we propose a novel Oral-Anatomical Knowledge-informed Semi-Supervised Learning (OAK-SSL) model for 3D CBCT image segmentation and lesion detection. The uniqueness of OAK-SSL is its capability of integrating qualitative oral-anatomical knowledge of plausible lesion locations into the deep learning design. Specifically, the unique design of OAK-SSL includes three key elements, including transformation of qualitative knowledge into quantitative representation, knowledge-informed dual-task learning architecture, and knowledge-informed semi-supervised loss function. We apply OAK-SSL to a real-world dataset, focusing on segmenting CBCT images that contain small lesions. This task is inherently challenging yet holds significant clinical value as treating lesions at their early stages lead to excellent prognosis. OAK-SSL demonstrated significantly better performance than a range of existing methods. Note to Practitioners-This study tackles the challenges arising from a limited amount of labeled data due to the time-consuming manual segmentation of 3D dental cone beam computed tomography (CBCT) images. The scarcity of labeled data often impedes AI models from accurately segmenting periapical lesions. To overcome this, we introduce a novel semi-supervised learning algorithm that integrates the oral-anatomical knowledge about lesion location for 3D CBCT image segmentation. Our method effectively segments periapical lesions, including even small-sized periapical lesions, without solely relying on labeled data. The proposed method offers two significant benefits to clinicians. First, it reduces the necessity for large amounts of labeled data, particularly easing the burden of manually segmenting early-stage periapical lesions. Second, it helps reduce intra- and inter-observer disagreements and human errors by providing consistent and automated segmentation maps. These benefits not only simplify the segmentation process in dental imaging but also improve its reliability. As a result, our automated algorithm makes it easier and more trustworthy for practitioners. However, practitioners should be aware that the effectiveness of our method relies on the assumption of consistency between labeled and unlabeled data. When applying this method, it is crucial to carefully consider the characteristics of the unlabeled dataset. Significant differences in image quality, patient demographics, or acquisition parameters between labeled and unlabeled data might affect model performance.
