IEEE Journal of Biomedical and Health Informatics最新文献

Fetal crown-rump length (CRL) is one of the most accurate method for estimating gestational age in early pregnancy. Typically, the process of manual CRL measurement by physicians is cumbersome, prone to errors due to fetal position, and susceptible to inter-observer variability. To provide an accurate, real-time, and reliable fetal CRL measurement solution, we propose FastCRL that utilizes key landmarks detection for efficient CRL measurements and fetal position estimation. Specifically, fast and lightweight network blocks are employed for both the encoder and decoder. By outputting low-resolution heatmaps and axial offset maps of key landmarks, we achieve a balance between high accuracy and fast inference speed. A novel Lightweight Adaptive Fourier Transform (LAFT) module is introduced to globally filter noise in ultrasound images and enhance the features required for landmark prediction. Additionally, the challenge of evaluating fetal position flexion and extension is effectively addressed by analyzing the angles between key landmarks on the fetal head, buttocks, and neck. The experimental results on our dataset indicate that our method for determining fetal position is both objective and efficient. FastCRL achieves a performance level consistent with the average human expert. In terms of measuring CRL, FastCRL achieved an error rate of less than 3% in 99.1% of measurements with 32 ms latency, significantly outperforming other baselines and demonstrating substantial potential for clinical application.

Breast cancer is one of the cancers of deep concern worldwide, and the molecular subtype of breast cancer is significant for patients' treatment selection, and prognosis judgment. The application of multi-sequence MRI technology provides a new non-invasive companion diagnostic method for molecular subtypes of breast cancer, which can more accurately assess the vascular status of tumors and reveal fine structures. However, providing interpretable classification results remains a challenge. Recently, although many convolutional neural network (CNN) methods and fine-grained classification methods based on MRI inputs have been proposed. However, most of these methods operate in a 'black-box' without a detailed explanation of the intermediate processes, resulting in a lack of interpretability of the breast cancer classification process. To address this problem, our study proposes a multi-sequence MRI-based hierarchical expert diagnostic method for the molecular subtype of breast cancer. With the strong differentiation module, this method first identifies enhanced features in breast tumors, ensuring that the subsequent classification process is precisely focused on the lesion features. In addition, inspired by the codiagnosis of multiple experts in clinical diagnosis, we set up a mechanism of collaborative diagnostic corrective learning by hierarchical experts to provide an interpretable classification process. Compared with previous studies, the framework learns features with a strong distinguishing ability for breast tumor classification. Specifically, multiple experts corrected each other's learning to give more accurate and interpretable classification results, significantly improving clinical diagnosis's practical value. We conducted extensive experiments on a breast dataset and compared it quantitatively with other methods, and we achieved the best performance in terms of accuracy (0.889) and F1 Score (0.893).We make the code public on GitHub: https://github.com/yanfangHao/HED.

Magnetoencephalography (MEG) is a vital non-invasive tool for epilepsy analysis, as it captures high-resolution signals that reflect changes in brain activity over time. The automated detection of epileptic spikes within these signals can significantly reduce the labor and time required for manual annotation of MEG recording data, thereby aiding clinicians in identifying epileptogenic foci and evaluating treatment prognosis. Research in this domain often utilizes the raw, multi-channel signals from MEG scans for spike detection, commonly neglecting the multi-channel spiking patterns from spatially adjacent channels. Moreover, epileptic spikes share considerable morphological similarities with artifact signals within the recordings, posing a challenge for models to differentiate between the two. In this paper, we introduce a multimodal fusion framework that addresses these two challenges collectively. Instead of relying solely on the signal recordings, our framework also mines knowledge from their corresponding topography-map images, which encapsulate the spatial context and amplitude distribution of the input signals. To facilitate more effective data fusion, we present a novel multimodal feature fusion technique called CrossConvPyramid, built upon a convolutional pyramid architecture augmented by an attention mechanism. It initially employs cross-attention and a convolutional pyramid to encode inter-modal correlations within the intermediate features extracted by individual unimodal networks. Subsequently, it utilizes a self-attention mechanism to refine and select the most salient features from both inter-modal and unimodal features, specifically tailored for the spike classification task. Our method achieved the average F1 scores of 92.88% and 95.23% across two distinct real-world MEG datasets from separate centers, respectively outperforming the current state-of-the-art by 2.31% and 0.88%. We plan to release the code on GitHub later.

Predicting drug-target binding affinity is critical for drug discovery, as it helps identify promising drug candidates and predict their effectiveness. Recent advancements in deep learning have made significant progress in tackling this task. However, existing methods heavily rely on training data, and their performance is often limited when predicting binding affinities for new drugs and targets. To address this challenge, we propose a novel Generalized Feature Learning (GFLearn) model for drug-target binding affinity prediction. By integrating Graph Neural Networks (GNNs) with a self-supervised invariant feature learning module, our GFLearn model can extract robust and highly generalizable features from both drugs and targets, significantly enhancing prediction performance. This innovation enables the model to effectively predict binding affinities for previously unseen drugs or targets, while also mitigates the common issue of prediction performance degrading due to shifts in data distribution. Extensive experiments were conducted on two diverse datasets across three challenging scenarios: new drugs, new targets, and combinations of both. Comparisons with state-of-the-art methods demonstrated that our GFLearn model consistently outperformed others, showcasing its robustness across various prediction tasks. Additionally, cross-dataset evaluations and noise perturbation experiments further validated the model's generalizability across different data distributions. Case studies on two drug-target pairs, Canertinib-PIK3C2G and MLN8054-FLT1, provided further evidence of GFLearn's ability to make accurate binding affinity predictions, offering valuable insights for drug screening and repurposing efforts.

Cancer is driven by specific mutations known as cancer driver genes, whose identification is crucial for advancing cancer therapy. Although many computational methods have been proposed with this purpose, most provide a single driver gene list ignoring the high heterogeneity of drivers across patients in cohort. Besides, they often fail to capture the higher-order interactions among genes at the patient level. Here we introduce a novel method PITCH to prioritize personalized cancer driver genes by assessing the higher-order propagation dynamics among genes. PITCH constructs a patient-specific hypergraph model that represents higher-order interactions well-characterized in signaling pathways, enabling a more comprehensive assessment of gene influence in cancer development. PITCH does not require paired case-control data, simplifying its application in clinical practice. We evaluated our approach using data from four different types of cancers, demonstrating its superior performance in identifying cancer driver genes compared to existing methods. Importantly, PITCH is shown to identify both common and rare drivers. The results were validated against well-studied cancer gene databases, confirming the accuracy of the identified drivers. Additionally, most of PITCH-identified personalized driver genes were actionable and druggable for most patients, offering significant potential for guiding personalized treatment strategies. Our approach represents a significant advancement in the field of cancer driver genes discovery, providing a powerful tool for the precise identification of therapeutic targets in cancer research.

Dynamic brain networks play a pivotal role in diagnosing brain disorders by capturing temporal changes in brain activity and connectivity. Previous methods often rely on sliding-window approaches for constructing these networks using fMRI data. However, these methods face two key limitations: a fixed temporal length that inadequately captures brain activity dynamics and a global spatial scope that introduces noise and reduces sensitivity to localized dysfunctions. These challenges can lead to inaccurate brain network representations and potential misdiagnoses.To address these challenges, we propose BrainDGT, a dynamic Graph Transformer model designed to enhance the construction and analysis of dynamic brain networks for more accurate diagnosis of brain disorders. BrainDGT leverages adaptive brain states by deconvolving the Hemodynamic Response Function (HRF) within individual functional brain modules to generate dynamic graphs, addressing the limitations of fixed temporal length and global spatial scope. The model learns spatio-temporal local features through attention mechanisms within these graphs and captures global interactions across modules using adaptive fusion. This dual-level integration enhances the model's ability to analyze complex brain connectivity patterns. We validate BrainDGT's effectiveness through classification experiments on three fMRI datasets (ADNI, PPMI, and ABIDE), where it outperforms state-of-the-art methods. By enabling adaptive, localized analysis of dynamic brain networks, BrainDGT advances neuroimaging and supports the development of more precise diagnostic and treatment strategies in biomedical research.

Cognitive control dysregulation is nearly universal across disorders, including major depressive disorder (MDD). Achieving comparable response rates to medication, the transcranial magnetic stimulation (TMS) mechanism and its effect on cognitive control have not been well understood yet. This paper investigates the predictive capability of the clinical response to TMS treatment using 34 cognitive variables measured from TMS treatment of 22 MDD subjects over an eight-week period. We employ a novel brain-inspired computing paradigm, hyperdimensional computing (HDC), to classify the effectiveness of TMS using leave-one-subject-out cross-validation (LOSOCV). Four performance metrics-accuracy, sensitivity, specificity and AUC-are used, with AUC being the primary metric. Experimental results reveal that: i). Although SVM outperforms HDC in terms of accuracy, HDC achieves an AUC of 0.82, surpassing SVM by 0.07. ii). The optimal performance for both classifiers is obtained with feature selection using SelectKBest. iii) Among the top features selected by SelectKBest for the two classifiers, ws_MedRT (median rate for the Websurf task) shows a more distinguishable distribution between clinical responses ("1") and no clinical responses ("0"). In conclusion, these results highlight the potential of HDC for predicting clinical responses to TMS and underscore the importance of feature selection in improving classification performance.

High-quality motion estimation is essential in ultrasound elastography (USE) for evaluating tissue mechanical properties and detecting abnormalities. Traditional methods, such as speckle tracking and regularized optimization, face challenges including noise, over-smoothing of displacements, and prolonged runtimes. Recent efforts have explored optical flow-based convolutional neural networks (CNNs). However, current approaches experience at least one of the following limitations: 1) reliance on tissue incompressibility assumption, which compromises data fidelity and can introduce large errors; 2) dependence on ground truth displacement data for supervised CNN methods; 3) use of a regularizer not aligned with tissue physics by relying only on first-order displacement derivatives; 4) use of a L2-norm regularizer that over-smoothes motion estimates; and 5) substantially large sampling size, increasing computational and memory demands, especially for classical methods. In this paper, we develop RAFT-USENet, a physics-informed, unsupervised neural network to estimate both axial and lateral displacements. We design RAFT-USENet by substantially modifying optical flow RAFT network to adapt it to high-frequency USE data. Extensive validation using simulation, phantom and in vivo data demonstrates that RAFT-USENet significantly improves motion estimation performance compared to recent classical and CNN methods. The normalized cross-correlation between pre- and warped post-deformation USE data using RAFT-USENet is estimated as 0.94, 0.88, and 0.82 in simulation, breast phantom and in vivo datasets, respectively, while corresponding comparative methods ranges were found 0.79-0.88, 0.76-0.85, and 0.69-0.81. Additionally, RAFT-USENet reduced computational time by 1.5-150 times compared to existing methods. These results suggest that RAFT-USENet may be a potentially useful reliable and accurate tool for clinical elasticity imaging applications.

mRNA subcellular localization is a prevalent and essential mechanism that precisely regulates protein translation and significantly impacts various cellular processes. mRNA subcellular localization has advanced the understanding of mRNA function, yet existing methods face limitations, including imbalanced data, suboptimal model performance, and inadequate generalization, particularly in multi-label localization scenarios where solutions are scarce. This study introduces MBSCLoc, a predictor for mRNA multi-label subcellular localization. MBSCLoc predicts mRNA locations across multiple cellular compartments simultaneously, overcoming challenges like single-location prediction, incomplete feature extraction, and imbalanced data. MBSCLoc leverages UTR-LM model for feature extraction, followed by multi-class contrastive representation learning and Clustering Balanced Subspace Partitioning to construct balanced subspaces. It then optimizes sample distribution to tackle severe data imbalance and uses multiple XGBoost classifiers, integrated through voting, to enhance accuracy and generalization. Five-fold cross-validation and independent testing results show that MBSCLoc significantly outperforms other methods. Additionally, MBSCLoc offers superior pixel-level interpretability, strongly supporting mRNA multi-label subcellular localization research. Crucially, the importance of the 5' UTR and 3' UTR regions has been preliminarily confirmed using traditional biological analysis and Tree-SHAP, with most mRNA sequences showing significant relevance in these regions, especially the 3' UTR where about 80% of specific sites reach peak significance. Concurrently, in order to facilitate the use of MBSCLoc by researchers, a freely accessible web has also been developed: http://www.mbscloc.com/.

As the elderly population grows, falling accidents become more frequent, and the need for fall-risk monitoring systems increases. Deep learning models for fallrisk movement detection neglect the connections between the terrain and fall-hazard movements. This issue can result in false alarms, particularly when a person encounters changing terrain. This work introduces a novel multisensor system that integrates terrain perception sensors with an inertial measurement unit (IMU) to monitor fall-risk on diverse terrains. Additionally, a dual-task learning (DTL) architecture that is based on a modified CNNLSTM model is implemented; it is used to determine fall-risk level and the terrain from sensor signals. Three fall-risk levels - "normal," "near-fall," and "fall" - are identified as being associated with "flat ground," "stepping up," and "stepping down" terrains. Ten young subjects performed 16 activities on flat and stepping terrains in a laboratory setting, and ten elderly individuals were recruited to perform four activities in the hospital. The accuracies of classification of fall-risk levels and terrains by the proposed system are 97.6% and 95.2%, respectively. The system detects pre-impact fall movements, with a fall prediction accuracy of 97.7% and an average lead time of 329ms for fall trials, revealing the model's effectiveness. The overall monitoring accuracy for elderly individuals is 99.8%, confirming the robustness of the proposed system. This work discusses the impact of sensor type and the model architecture of DTL on the classification of fall-risk levels across various terrains. The results demonstrate that the proposed method is reliable for monitoring the risk of falling.