IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

BACKGROUND: Access to palliative care is a key quality metric which most healthcare organizations strive to improve. The primary challenges to increasing palliative care access are a combination of physicians over-estimating patient prognoses, and a shortage of palliative staff in general. This, in combination with treatment inertia can result in a mismatch between patient wishes, and their actual care towards the end of life. METHODS: In this work, we address this problem, with Institutional Review Board approval, using machine learning and Electronic Health Record (EHR) data of patients. We train a Deep Neural Network model on the EHR data of patients from previous years, to predict mortality of patients within the next 3-12 month period. This prediction is used as a proxy decision for identifying patients who could benefit from palliative care. RESULTS: The EHR data of all admitted patients are evaluated every night by this algorithm, and the palliative care team is automatically notified of the list of patients with a positive prediction. In addition, we present a novel technique for decision interpretation, using which we provide explanations for the model's predictions. CONCLUSION: The automatic screening and notification saves the palliative care team the burden of time consuming chart reviews of all patients, and allows them to take a proactive approach in reaching out to such patients rather then relying on referrals from the treating physicians.

Recently, some pioneering works have preferred applying more complex modules to improve segmentation performances. However, it is not friendly for actual clinical environments due to limited computing resources. To address this challenge, we propose a light-weight model to achieve competitive performances for skin lesion segmentation at the lowest cost of parameters and computational complexity so far. Briefly, we propose four modules: (1) DGA consists of dilated convolution and gated attention mechanisms to extract global and local feature information; (2) IEA, which is based on external attention to characterize the overall datasets and enhance the connection between samples; (3) CAB is composed of 1D convolution and fully connected layers to perform a global and local fusion of multi-stage features to generate attention maps at channel axis; (4) SAB, which operates on multi-stage features by a shared 2D convolution to generate attention maps at spatial axis. We combine four modules with our U-shape architecture and obtain a light-weight medical image segmentation model dubbed as MALUNet. Compared with UNet, our model improves the mIoU and DSC metrics by 2.39% and 1.49%, respectively, with a 44x and 166x reduction in the number of parameters and computational complexity. In addition, we conduct comparison experiments on two skin lesion segmentation datasets (ISIC2017 and ISIC2018). Experimental results show that our model achieves state-of-the-art in balancing the number of parameters, computational complexity and segmentation performances. Code is available at https://github.com/JCruan519/MALUNet.

Malaria is a major global health threat. The standard way of diagnosing malaria is by visually examining blood smears for parasite-infected red blood cells under the microscope by qualified technicians. This method is inefficient and the diagnosis depends on the experience and the knowledge of the person doing the examination. Automatic image recognition technologies based on machine learning have been applied to malaria blood smears for diagnosis before. However, the practical performance has not been sufficient so far. This study proposes a new and robust machine learning model based on a convolutional neural network (CNN) to automatically classify single cells in thin blood smears on standard microscope slides as either infected or uninfected. In a ten-fold cross-validation based on 27,578 single cell images, the average accuracy of our new 16-layer CNN model is 97.37%. A transfer learning model only achieves 91.99% on the same images. The CNN model shows superiority over the transfer learning model in all performance indicators such as sensitivity (96.99% vs 89.00%), specificity (97.75% vs 94.98%), precision (97.73% vs 95.12%), F1 score (97.36% vs 90.24%), and Matthews correlation coefficient (94.75% vs 85.25%).

Automatic emotion recognition based on multi-channel neurophysiological signals, as a challenging pattern recognition task, is becoming an important computer-aided method for emotional disorder diagnoses in neurology and psychiatry. Traditional approaches require designing and extracting a range of features from single or multiple channel signals based on extensive domain knowledge. This may be an obstacle for non-domain experts. Moreover, traditional feature fusion method can not fully utilize correlation information between different channels. In this paper, we propose a preprocessing method that encapsulates the multi-channel neurophysiological signals into grid-like frames through wavelet and scalogram transform. We further design a hybrid deep learning model that combines the `Convolutional Neural Network (CNN)' and `Recurrent Neural Network (RNN)', for extracting task-related features, mining inter-channel correlation and incorporating contextual information from those frames. Experiments are carried out, in a trial-level emotion recognition task, on the DEAP benchmarking dataset. Our results demonstrate the effectiveness of the proposed methods, with respect to the emotional dimensions of Valence and Arousal.

Detection of Alzheimer's Disease (AD) from neuroimaging data such as MRI through machine learning have been a subject of intense research in recent years. Recent success of deep learning in computer vision have progressed such research further. However, common limitations with such algorithms are reliance on a large number of training images, and requirement of careful optimization of the architecture of deep networks. In this paper, we attempt solving these issues with transfer learning, where state-of-the-art architectures such as VGG and Inception are initialized with pre-trained weights from large benchmark datasets consisting of natural images, and the fully-connected layer is re-trained with only a small number of MRI images. We employ image entropy to select the most informative slices for training. Through experimentation on the OASIS MRI dataset, we show that with training size almost 10 times smaller than the state-of-the-art, we reach comparable or even better performance than current deep-learning based methods.

Traditional Cox proportional hazard model for survival analysis are based on structured features like patients' sex, smoke years, BMI, etc. With the development of medical imaging technology, more and more unstructured medical images are available for diagnosis, treatment and survival analysis. Traditional survival models utilize these unstructured images by extracting human-designed features from them. However, we argue that those hand-crafted features have limited abilities in representing highly abstract information. In this paper, we for the first time develop a deep convolutional neural network for survival analysis (DeepConvSurv) with pathological images. The deep layers in our model could represent more abstract information compared with hand-crafted features from the images. Hence, it will improve the survival prediction performance. From our extensive experiments on the National Lung Screening Trial (NLST) lung cancer data, we show that the proposed DeepConvSurv model improves significantly compared with four state-of-the-art methods.

Computer aid technology is widely applied in decision-making and outcome assessment of healthcare delivery, in which modeling knowledge and expert experience is technically important. However, the conventional rule-based models are incapable of capturing the underlying knowledge because they are incapable of simulating the complexity of human brains and highly rely on feature representation of problem domains. Thus we attempt to apply a deep model to overcome this weakness. The deep model can simulate the thinking procedure of human and combine feature representation and learning in a unified model. A modified version of convolutional deep belief networks is used as an effective training method for large-scale data sets. Then it is tested by two instances: a dataset on hypertension retrieved from a HIS system, and a dataset on Chinese medical diagnosis and treatment prescription from a manual converted electronic medical record (EMR) database. The experimental results indicate that the proposed deep model is able to reveal previously unknown concepts and performs much better than the conventional shallow models.

In this paper, we demonstrate the application of Mask-RCNN, the state-of-the-art convolutional neural network algorithm for object detection and segmentation to the oral pathology domain. Mask-RCNN was originally developed for object detection, and object instance segmentation of natural images. With this experiment, we show that Mask-RCNN can also be used in a very specialized area such as oral pathology. While the number of oral diseases are numerous and varied in the form of Thrush, Leukoplakia, Lichenplanus, etc., we limited our scope to the detection and instance segmentation of two of the most commonly occurring conditions, herpes labialis (commonly referred to as “cold sore“ and aphthous ulcer (commonly referred to as “canker sore“ This paper aims at detecting and segmenting cold sores and canker sores only. As always, no computer based detection system can be 100% reliable. An accurate diagnosis by a trained health care professional is necessary since several conditions of the mouth including oral cancer may mimic canker sores.

In this paper1, we proposed an explainable deep neural networks (DNN)-based method for automatic detection of COVID-19 symptoms from chest radiography (CXR) images, which we call ‘DeepCOVIDExplainer’. We used 15,959 CXR images of 15,854 patients, covering normal, pneumonia, and COVID-19 cases. CXR images are first comprehensively preprocessed and augmented before classifying with a neural ensemble method, followed by highlighting class-discriminating regions using gradient-guided class activation maps (Grad-CAM ++) and layer-wise relevance propagation (LRP). Further, we provide human-interpretable explanations for the diagnosis. Evaluation results show that our approach can identify COVID-19 cases with a positive predictive value (PPV) of 91.6%, 92.45%, and 96.12%, respectively for normal, pneumonia, and COVID-19 cases, respectively, outperforming recent approaches.1Read longer version of this paper: https://arxiv.org/pdf/2004.04582.pdf

Transcription factors (TFs) are macromolecules that bind to cis-regulatory specific sub-regions of DNA promoters and initiate transcription. Finding the exact location of these binding sites (aka motifs) is important in a variety of domains such as drug design and development. To address this need, several in vivo and in vitro techniques have been developed so far that try to characterize and predict the binding specificity of a protein to different DNA loci. The major problem with these techniques is that they are not accurate enough in prediction of the binding affinity and characterization of the corresponding motifs. As a result, downstream analysis is required to uncover the locations where proteins of interest bind. Here, we propose DeeperBind, a long short term recurrent convolutional network for prediction of protein binding specificities with respect to DNA probes. DeeperBind can model the positional dynamics of probe sequences and hence reckons with the contributions made by individual sub-regions in DNA sequences, in an effective way. Moreover, it can be trained and tested on datasets containing varying-length sequences. We apply our pipeline to the datasets derived from protein binding microarrays (PBMs), an in-vitro high-throughput technology for quantification of protein-DNA binding preferences, and present promising results. To the best of our knowledge, this is the most accurate pipeline that can predict binding specificities of DNA sequences from the data produced by high-throughput technologies through utilization of the power of deep learning for feature generation and positional dynamics modeling.

Recently, researchers have begun to experiment with deep learning-based methods for detecting major depressive disorder (MDD) using electroencephalogram (EEG) signals in search of a more objective means of diagnosis. However, existing spatiotemporal feature extraction methods only consider the functional correlation between multiple electrodes and temporal correlation of EEG signals, ignoring the spatial position connection information between electrodes and the continuity between time windows, which reduces the model's feature extraction capabilities. To address this issue, a Spatiotemporal fused network for MDD detection with Electrode spatial Topology and adjacent TIME-window transition information (SETTIME) is proposed in this study. SET-TIME is composed by a common feature extractor, a secondary time-correlation feature extractor, and a domain adaptation (DA) module, in which the former extractor is used to obtain the temporal and spatial features, and the latter extractor can mine the correlation between multiple time windows, and the DA module is adopted to enhance cross-subject detection capability. The experimental results of 10-fold cross-validation show that the proposed SETTIME method outperforms the state-of-the-art (SOTA) method by achieving MDD detection accuracies of 92.00% and 94.00% on the public datasets PRED+CT and MODMA, respectively. Ablation experiments demonstrate the effectiveness of the multiple modules in SET-TIME, which assist in MDD detection by exploring the intrinsic spatiotemporal information of EEG signals.

High-fidelity 3D anatomical reconstruction is essential for clinical applications, yet converting voxel-based segmentation into smooth and watertight meshes remains challenging due to noise, incompleteness, and shape variability. We propose Med-PU, a knowledge-driven framework that integrates volumetric segmentation with deep point cloud upsampling to reconstruct high-quality anatomical surfaces. Instead of manual landmarking or PCA-based statistical models, Med-PU learns an implicit anatomical prior from large-scale 3D medical shapes, enabling dense and topology-consistent reconstruction from sparse segmentation outputs. Experiments on pelvic CT datasets (Med-ShapePelvic for training and Pelvic1k for validation) show that Med-PU consistently improves geometric accuracy and surface fidelity over state-of-the-art upsampling methods, demonstrating robustness across input densities and potential generalization to other anatomical regions.

Drug combinations are essential in cancer therapy, leveraging synergistic drug-drug interactions (DDI) to enhance efficacy and combat resistance. However, the vast combinatorial space makes experimental screening impractical, and existing computational models struggle to capture the complex, bidirectional nature of DDIs, often relying on independent drug encoding or simplistic fusion strategies. To address this, we propose DPASyn, a novel drug synergy prediction framework featuring a dual-attention mechanism and Precision-Aware Quantization (PAQ). The dual-attention architecture jointly models intra-drug structures and inter-drug interactions via shared projections and cross-drug attention, enabling biologically plausible synergy modeling. Our PAQ strategy dynamically optimizes numerical precision during training based on feature sensitivity—reducing memory usage by 40% and accelerating training threefold without sacrificing accuracy. With LayerNorm-stabilized residual connections for stability, DPASyn outperforms seven state-of-the-art methods on the O'Neil dataset (13,243 combinations) [1] and supports full-batch processing of up to 256 graphs on a single GPU—setting a new standard for efficient and expressive drug synergy prediction. The data and source code are available at https://github.com/Echo-Nie/DPASyn.

Positron Emission Tomography (PET) is crucial in medicine, but its clinical use is limited due to high signal-to-noise ratio doses increasing radiation exposure. Lowering doses increases Poisson noise, which current denoising methods fail to handle, causing distortions and artifacts. We propose a Poisson Consistent U-Net (PC-UNet) model with a new Poisson Variance and Mean Consistency Loss (PVMC-Loss) that incorporates physical data to improve image fidelity. PVMC- Loss is statistically unbiased in variance and gradient adaptation, acting as a Generalized Method of Moments implementation, offering robustness to minor data mismatches. Tests on PET datasets show PC-UNet improves physical consistency and image fidelity, proving its ability to integrate physical information effectively.

The reconstruction of visual information from brain activity fosters interdisciplinary integration between neuroscience and computer vision. However, existing methods still face challenges in accurately recovering highly complex visual stimuli. This difficulty stems from the characteristics of natural scenes: low-level features exhibit heterogeneity, while high-level features show semantic entanglement due to contextual overlaps. Inspired by the hierarchical representation theory of the visual cortex, we propose the HAVIR model, which separates the visual cortex into two hierarchical regions and extracts distinct features from each. Specifically, the Structural Generator extracts structural information from spatial processing voxels and converts it into latent diffusion priors, while the Semantic Extractor converts semantic processing voxels into CLIP embeddings. These components are integrated via the Versatile Diffusion model to synthesize the final image. Experimental results demonstrate that HAVIR enhances both the structural and semantic quality of reconstructions, even in complex scenes, and outperforms existing models.

Functional magnetic resonance imaging (fMRI) is a powerful tool for probing brain function, yet reliable clinical diagnosis is hampered by low signal-to-noise ratios, inter-subject variability, and the limited frequency awareness of prevailing CNN- and Transformer-based models. Moreover, most fMRI datasets lack textual annotations that could contextualize regional activation and connectivity patterns. We introduce RTGMFF, a framework that unifies automatic ROI-level text generation with multimodal feature fusion for brain-disorder diagnosis. RTGMFF consists of three components: (i) ROI-driven fMRI text generation deterministically condenses each subject's activation, connectivity, age, and sex into reproducible text tokens; (ii) Hybrid frequency-spatial encoder fuses a hierarchical waveletmamba branch with a cross-scale Transformer encoder to capture frequency-domain structure alongside long-range spatial dependencies; and (iii) Adaptive semantic alignment module embeds the ROI token sequence and visual features in a shared space, using a regularized cosine-similarity loss to narrow the modality gap. Extensive experiments on the ADHD-200 and ABIDE benchmarks show that RTGMFF surpasses current methods in diagnostic accuracy, achieving notable gains in sensitivity, specificity, and area under the ROC curve. Code is available at https://github.com/BeistMedAI/RTGMFF.

Accurate recognition of driver drowsiness is essential for preventing accidents caused by reduced alertness, yet cross-subject electroencephalography (EEG) analysis remains hindered by noisy and individually variable brain activity. Existing methods lack the capacity to discern the contribution of each electrode, leading to suboptimal feature selection and degraded recognition accuracy. Here we introduce an auditory-inspired electrode attention (AEA) mechanism that emulates human selective hearing to isolate the most informative drowsinessrelated electrodes from noisy EEG signals. AEA integrates a fluctuation feature extraction module (FFEM) and an electrode value estimation module (EVEM), supported by an electrode contribution (EC) metric that quantifies electrode importance. FFEM captures spatiotemporal fluctuation dynamics, while EVEM models their global correlations, jointly enhancing robustness to individual variability and motion artifacts. When combined with conventional backbones, AEA achieves accuracies of 83.31 % and 80.45 % on the SADT and SEED-VIG datasets, respectively-surpassing state-of-the-art methods by up to 4.96 % in accuracy and$\mathbf{1 2. 5 5 \%}$in F1-score. Code available here.

The analysis of unstructured text, such as clinical notes and public health reports, is a fundamental challenge in biomedical and health informatics (BHI). Occupational health informatics, a vital sub-domain of BHI, relies on analyzing injury narratives to prevent workplace accidents, which constitute a significant public health burden. While Large Language Models (LLMs) offer a promising direction, their practical deployment in health-related applications is often hindered by the prohibitive computational costs of fine-tuning and the unreliability of prompt-based methods. To address these limitations, we propose an efficient, fine-tune-free, and lightweight fusion framework for robust text classification. Our framework introduces a primary-auxiliary architecture that harnesses diverse semantic embeddings from multiple off-the-shelf LLMs. By modeling the second-order interactions between these embeddings via a cooccurrence mechanism, our approach generates a powerful, composite feature representation for a simple downstream classifier. We empirically validate our framework on a public benchmark of occupational injury reports from the U.S. Occupational Safety and Health Administration (OSHA), where the framework achieves a state-of-the-art weighted-F1 score of$\mathbf{6 8. 2 0 \%}$. Notably, our framework outperforms a purpose-built, fine-tuned LLM of the same scale while demonstrating a 4.4-fold improvement in computational efficiency. The code is publicly available at: https://github.com/nxcc-lab/LLM-Fusion-Framework.

Brain decoding is a key neuroscience field that reconstructs the visual stimuli from brain activity with fMRI, which helps illuminate how the brain represents the world. fMRI-to-image reconstruction has achieved impressive progress by leveraging diffusion models. However, brain signals infused with prior knowledge and associations exhibit a significant information asymmetry when compared to raw visual features, still posing challenges for decoding fMRI representations under the supervision of images. Consequently, the reconstructed images often lack fine-grained visual fidelity, such as missing attributes and distorted spatial relationships. To tackle this challenge, we propose BrainCognizer, a novel brain decoding model inspired by human visual cognition, which explores multilevel semantics and correlations without fine-tuning of generative models. Specifically, BrainCognizer introduces two modules: the Cognitive Integration Module which incorporates prior human knowledge to extract hierarchical region semantics; and the Cognitive Correlation Module which captures contextual semantic relationships across regions. Incorporating these two modules enhances intra-region semantic consistency and maintains interregion contextual associations, thereby facilitating fine-grained brain decoding. Moreover, we quantitatively interpret our components from a neuroscience perspective and analyze the associations between different visual patterns and brain functions. Extensive quantitative and qualitative experiments demonstrate that BrainCognizer outperforms state-of-the-art approaches on multiple evaluation metrics. Our code is released publicly at https://github.com/Grace160/BrainCognizer.

Understanding the underlying mechanisms of TCR-epitope interactions is crucial for studying the adaptive immune system and promoting the field of immunotherapy. Given the high cost of traditional experimental methods, it is urgent to develop computational methods to predict TCRepitope binding. With the advancement of experimental technology, an increasing number of TCR-epitope binding pairs have been archived in public databases, creating opportunities for the advancement of computational methods. In this study, we propose a novel framework called MKDBAN-TEI for predicting TCR-epitope binding. We encode TCR and epitope sequences using a learnable residue embedding matrix and employ CNN layers to extract features. An interpretable bilinear attention network is then used to capture the interaction patterns between TCR and epitope. To improve the model's performance and generalization capability, we introduce a multi-level knowledge distillation framework: first, we cluster epitopes in the training set based on sequence similarity to define distinct domains; second, for each domain, we integrate three types of protein sequence features (protein language embeddings, physicochemical information, and evolutionary information) to train domain-specific teacher models via internal multi-feature knowledge distillation, capturing domain-specific binding patterns; finally, we distill knowledge from all domain-specific teachers to a universal student model through inter-domain knowledge distillation, enhancing generalization to unseen epitopes. Compared to several state-of-the-art models, MKDBAN-TEI demonstrates superior performance and generalization capability. Further experiments illustrate the effectiveness of the model in realworld scenarios. Visualizing the attention maps learned by MKDBAN-TEI provides new insights into TCR-epitope interactions. The code is available at: https://github.com/X/MKDBAN-TEI