Pulmonary Arterial Wedge Pressure (PAWP) is an essential cardiovascular hemodynamics marker to detect heart failure. In clinical practice, Right Heart Catheterization is considered a gold standard for assessing cardiac hemodynamics while non-invasive methods are often needed to screen high-risk patients from a large population. In this paper, we propose a multimodal learning pipeline to predict PAWP marker. We utilize complementary information from Cardiac Magnetic Resonance Imaging (CMR) scans (short-axis and four-chamber) and Electronic Health Records (EHRs). We extract spatio-temporal features from CMR scans using tensor-based learning. We propose a graph attention network to select important EHR features for prediction, where we model subjects as graph nodes and feature relationships as graph edges using the attention mechanism. We design four feature fusion strategies: early, intermediate, late, and hybrid fusion. With a linear classifier and linear fusion strategies, our pipeline is interpretable. We validate our pipeline on a large dataset of ${2},{641}$ subjects from our ASPIRE registry. The comparative study against state-of-the-art methods confirms the superiority of our pipeline. The decision curve analysis further validates that our pipeline can be applied to screen a large population. The code is available at https://github.com/prasunc/hemodynamics.

Unsupervised Domain Adaptation (UDA) can transfer knowledge from labeled source data to unlabeled target data of the same categories. However, UDA for first-person video action recognition is an under-explored problem, with a lack of benchmark datasets and limited consideration of first-person video characteristics. Existing benchmark datasets provide videos with a single activity scene, e.g. kitchen, and similar global video statistics. However, multiple activity scenes and different global video statistics are still essential for developing robust UDA networks for real-world applications. To this end, we first introduce two first-person video domain adaptation datasets: ADL-7 and GTEA_KITCHEN-6. To the best of our knowledge, they are the first to provide multi-scene and cross-site settings for UDA problem on first-person video action recognition, promoting diversity. They provide five more domains based on the original three from existing datasets, enriching data for this area. They are also compatible with existing datasets, ensuring scalability. First-person videos have unique challenges, i.e. actions tend to occur in hand-object interaction areas. Therefore, networks paying more attention to such areas can benefit common feature learning in UDA. Attention mechanisms can endow networks with the ability to allocate resources adaptively for the important parts of the inputs and fade out the rest. Hence, we introduce channel-temporal attention modules to capture the channel-wise and temporal-wise relationships and model their inter-dependencies important to this characteristic. Moreover, we propose a Channel-Temporal Attention Network (CTAN) to integrate these modules into existing architectures. CTAN outperforms baselines on the new datasets and one existing dataset, EPIC-8.

Machine learning has been widely used to develop classification models for autism spectrum disorder (ASD) using neuroimaging data. Recently, studies have shifted towards using large multi-site neuroimaging datasets to boost the clinical applicability and statistical power of results. However, the classification performance is hindered by the heterogeneous nature of agglomerative datasets. In this paper, we propose new methods for multi-site autism classification using the Autism Brain Imaging Data Exchange (ABIDE) dataset. We firstly propose a new second-order measure of functional connectivity (FC) named as Tangent Pearson embedding to extract better features for classification. Then we assess the statistical dependence between acquisition sites and FC features, and take a domain adaptation approach to minimize the site dependence of FC features to improve classification. Our analysis shows that 1) statistical dependence between site and FC features is statistically significant at the 5% level, and 2) extracting second-order features from neuroimaging data and minimizing their site dependence can improve over state-of-the-art (SOTA) classification results, achieving a classification accuracy of 73%. The code is available at https://github.com/kundaMwiza/fMRI-site-adaptation.

PyKale is a Python library for Knowledge-aware machine learning from multiple sources of data to enable/accelerate interdisciplinary research. It embodies green machine learning principles to reduce repetitions/redundancy, reuse existing resources, and recycle learning models across areas. We propose a pipeline-based application programming interface (API) so all machine learning workflows follow a standardized six-step pipeline. PyKale focuses on leveraging knowledge from multiple sources for accurate and interpretable prediction, particularly multimodal learning and transfer learning. To be more accessible, it separates code and configurations to enable non-programmers to configure systems without coding. PyKale is officially part of the PyTorch ecosystem and includes interdisciplinary examples in bioinformatics, knowledge graph, image/video recognition, and medical imaging: https://pykale.github.io/.

Independent Component Analysis (ICA) is a fundamental method for Blind Source Separation (BSS). Classical ICA takes data matrix input formed by vector data. This paper focuses on ICA for BSS with third-order data tensor input formed by matrix data, such as 2D images. Two approaches exist for this problem. The first approach reshapes each matrix into a vector to apply classical ICA, with structural information lost. The second approach unfolds a data tensor into a data matrix along different modes to perform classical ICA mode-wise, which partially preserves structures but has strong or ill BSS assumptions. This paper proposes a third approach via RAndom Matrix ICA (RAMICA) modeling. RAMICA works on data tensor directly, without vectorization or unfolding, and preserves row or column structures under more general BSS assumptions. We develop the RAMICA model, algorithm, and related theories via defining new statistics for random matrices and new procedures for whitening and independent component estimation. We study the identifiability, higher-order extension, and relationships with existing methods. Experiments on both synthetic and real data show superior BSS performance of RAMICA over competing methods and offer insights on the trade-offs between different factors.

The increasing of public neuroimaging datasets opens a door to analyzing homogeneous human brain conditions across datasets by transfer learning (TL). However, neuroimaging data are high-dimensional, noisy, and with small sample sizes. It is challenging to learn a robust model for data across different cognitive experiments and subjects. A recent TL approach minimizes domain dependence to learn common cross-domain features, via the Hilbert-Schmidt Independence Criterion (HSIC). Inspired by this approach and the multi-source TL theory, we propose a Side Information Dependence Regularization (SIDeR) learning framework for TL in brain condition decoding. Specifically, SIDeR simultaneously minimizes the empirical risk and the statistical dependence on the domain side information, to reduce the theoretical generalization error bound. We construct 17 brain decoding TL tasks using public neuroimaging data for evaluation. Comprehensive experiments validate the superiority of SIDeR over ten competing methods, particularly an average improvement of 15.6% on the TL tasks with multi-source experiments.