Coronary artery disease (CAD) and valvular heart disorder (VHD) are major categories of cardiovascular disease (CVD), the leading cause of mortality and morbidity worldwide. CAD occurs due to plaque accumulation on the inner walls of the coronary arteries, restricting blood flow to the myocardium and potentially leading to heart attack or stroke. VHD refers to dysfunction in one or more heart valves, impairing blood flow between the heart's chambers or to other systemic organs. Due to the prevalence of CVD, there is a global need for an effective screening tool capable of detecting various CVDs on a mass scale. Both CAD and VHD alter the acoustic signature of phonocardiogram (PCG) signals, offering a non-invasive detection method. This study implements a multiclass classification model to differentiate between CAD, VHD, and normal heartbeats, collected from subjects using an innovative wearable multichannel PCG vest. Linear frequency cepstral coefficients (LFCCs), coupled with a support vector machine (SVM) using a radial basis function (RBF) kernel, achieved the highest multiclass subject-level accuracy of 81.53%, with a sensitivity of 85.20% for VHD and 81.47% for CAD. Additionally, a binary classification task between normal and abnormal (CAD and VHD together) heartbeats reported an accuracy of 82.01%. This is the first study to apply multiclass classification across different CVD categories using PCG signals collected with the same hardware.

Object tracking is a crucial component in the medical field, with great potential to enhance clinical workflows, particularly when integrated with augmented reality technologies. Accurate and reliable tracking systems can improve precision, usability, and operator feedback, facilitating innovative applications such as telemedicine, medical training, and robot-assisted procedures. This study evaluates the performance of FoundationPose, a neural network designed for six-degree-of-freedom pose estimation and real-time object tracking, in the context of ultrasound probe tracking. The RGB and depth images necessary for the network's operation were acquired using an Intel RealSense D435 3D camera. The feasibility and accuracy of FoundationPose were evaluated by analysing its ability to estimate both the translational and rotational components of the probe's pose. Experimental results demonstrated the network's ability to achieve mean errors of less than 6mm in distance estimation and under 1° in rotation tracking, with low sensitivity to the initialisation point. These findings confirm the potential of FoundationPose for real-time ultrasound probe tracking in controlled conditions. Future developments could focus on integrating this system with augmented reality platforms to provide real-time visual guidance and enhance clinical applications.Clinical relevance-This system enables precise real-time ultrasound probe tracking, enhancing procedural accuracy and supporting advanced clinical applications like augmented reality guidance and robotic-assisted interventions.

Cuffless blood pressure monitoring enables noninvasive, practical, and continuous cardiovascular disease management. This paper introduces an intelligent inertial blood pressure measurement device based on pulse transit time and ballistocardiography. Time-synchronized data from two accelerometers placed on the common carotid artery are fed to a lightweight on-device neural network for systolic and diastolic estimation. Platform-aware neural architecture search pre-trains the most accurate model that fits within the device from a multi-person dataset. Transfer learning personalizes the model using only 30 seconds of cuff calibration. The model achieves < ±2 mmHg systolic and diastolic error in preclinical trials on 18 patients, consumes <100 kB of memory, and provides 3 readings every second. The device is wireless, non-invasive, personalizable, and cuffless, communicating with patients using a smartphone interface. The framework enables continuous blood pressure monitoring in non-clinical settings by non-experts.

Ensemble Learning is a machine learning method that enhances overall predictive performance by combining multiple base learners. However, most current ensemble learning approaches employ average fusion methods, which overlook the consistency and diversity of individual model predictions and are unable to adaptively handle testing data. This paper introduces an Evidence Reasoning (ER) rule ensemble learning method that unifies model adaptation, uncertainty estimation, and confidence calibration within a single framework, thereby providing a more reliable model to aid physicians in decision-making. We evaluated our approach in predicting locoregional recurrence in Head and Neck Cancer (HNC). Compared to the previously proposed ERE, the ER-rule ensemble model achieved a 4.1% improvement in ACC.Clinical Relevance-This ER-rule ensemble model demonstrates a more reliable approach to predicting locoregional recurrence in head and neck cancer, enabling timely clinical intervention and potentially improving patient outcomes.

Early diagnosis of systemic sclerosis (SSc) is critical for early intervention and improved patient outcomes. This study explores the integration of near-infrared spectroscopy (NIRS) with deep learning for classification of SSc patients based on hand perfusion patterns. A probabilistic convolutional neural network (CNN) using MobileNetV2 with transfer learning was employed to analyze NIRS-derived oxygen saturation maps. The model achieved a test accuracy of 87.5%, demonstrating strong classification performance despite the limited data set. The Monte Carlo Dropout method was incorporated to evaluate the predictive uncertainty, providing valuable insight into the confidence of the model and its ability to detect potential out-of-distribution (OOD) inputs. The different confidence levels observed in the training, validation, and test datasets highlight the importance of uncertainty estimation in assessing model reliability and robustness. These results underscore the feasibility of deep learning-based NIRS analysis as a noninvasive and automated tool to detect microvascular dysfunction in SSc patients. Future work should focus on expanding the dataset, integrating multimodal imaging, and exploring advanced architectures to improve generalizability and clinical applicability.Clinical Relevance- Early detection of systemic sclerosis is essential for better outcomes. NIRS, combined with deep learning, offers a non-invasive, objective, and efficient tool for improving diagnosis and monitoring of microvascular dysfunction in systemic sclerosis patients.

The distributed neural interface demonstrates significant potential for effective disease treatments, where the power supply is a critical challenge to be addressed. The body-coupled power transfer (BCP) emerges as a flexible and efficient solution that allows electrode misalignment and supports longer transmission distances. However, the use of BCP to power fully implantable multiple implants has not been sufficiently explored. This paper utilizes electromagnetic simulations and practical experiments to observe the inter-effect of multi-nodes on their respective path gain. Two types of BCP (galvanic coupling and capacitive coupling) are studied, and their performance is compared across three implanted nodes. The results demonstrate that BCP can effectively power multiple implanted receivers with minimal mutual interference, which provides essential insights for its implementation in distributed neural interfaces.

Cardiopulmonary sounds contain a rich reservoir of vital and pathological information critical for clinical diagnosis. This paper presents a novel approach to cardiopulmonary data capturing with compressive sensing and reconstruction using a Convolutional Neural Network (CNN) based on the U-Net architecture. Applying traditional compressive sensing techniques to cardiopulmonary sounds presents several challenges. Cardiopulmonary sounds are inherently complex, with a substantial variation between captures. The traditional algorithms for compressive sensing rely on signal sparsity, whereas finding a sparse representation domain for cardiopulmonary sounds is a difficult task. Instead of finding a sparse domain manually, we propose training a convolutional encoder-decoder neural network for a pseudo-randomly undersampled set of signals without explicitly enforcing the sparsity concept. In this research, a CNN was trained for pseudo-randomly decimated input signals, evaluating a compression ratio of up to 30. The network is trained for respiratory sounds using the SPRSound dataset and for Phonocardiogram (PCG) signals using the CirCor Digiscope PCG dataset. Both these datasets have been evaluated for signal integrity after reconstruction and delivered promising results. The algorithm achieves reconstruction quality similar to that of previous research with a compression ratio three times higher than that of previous research applied to respiratory sounds. Since the principles of compressive sensing are applied in the sampling stage, the data compression requires no computation in the compression stage, and can therefore easily be implemented in low-cost edge devices.Clinical relevance- This work enables efficient compression of cardiopulmonary sounds, maintaining high signal integrity even at three times higher compression ratios than previous methods applied to respiratory sounds. It supports low-power, portable devices for real-time monitoring, improving accessibility for telemedicine and point-of-care diagnostics in respiratory and cardiovascular conditions.

As an advanced image argumentation approach, image generation technology offers a novel solution to the challenges of data scarcity and distribution imbalance in the medical field. However, the severe imbalance in the class distribution causes the networks to overfit to the head classes, while failing to adequately model the distribution of the tail class data during image generation, ultimately compromising the quality of the generated images. To solve this problem, we propose a Class Prototype-Driven Diffusion Model (CPDM) to improve class-conditional image synthesis on long-tailed chest X-ray images datasets. To fully extract the features of limited tail classes while avoiding overfitting to head classes, we introduce a Class Prototype Bank, which stores representative feature vectors of each class. Furthermore, by integrating cross-attention mechanisms between image features and class-specific prototypes, CPDM effectively captures fine-grained class features, enhancing both the realism and diversity of the generated images. Experiments show that our CPDM achieves the lowest FID=31.600 and highest IS=2.842, highlighting the effectiveness of CPDM in mitigating class imbalance and data scarcity in chest X-ray imaging. In downstream experiments, the classifier achieves a 17.22% improvement on the mAUC for 14 thoracic diseases when trained on a mixed dataset containing only 1% real images.

Auscultation remains a cornerstone of clinical practice, essential for both initial evaluation and continuous monitoring. Clinicians listen to the lung sounds and make a diagnosis by combining the patient's medical history and test results. Given this strong association, multitask learning (MTL) can offer a compelling framework to simultaneously model these relationships, integrating respiratory sound patterns with disease manifestations. While MTL has shown considerable promise in medical applications, a significant research gap remains in understanding the complex interplay between respiratory sounds, disease manifestations, and patient metadata attributes. This study investigates how integrating MTL with cutting-edge deep learning architectures can enhance both respiratory sound classification and disease diagnosis. Specifically, we extend recent findings regarding the beneficial impact of metadata on respiratory sound classification by evaluating its effectiveness within an MTL framework. Our comprehensive experiments reveal significant improvements in both lung sound classification and diagnostic performance when the stethoscope information is incorporated into the MTL architecture.Clinical relevance Our integrated MTL approach has immediate clinical applications in supporting medical professionals' diagnostic decisions, including lung sound classification to aid in detecting respiratory disorders, potentially reducing misdiagnosis rates and improving patient outcomes in respiratory care settings (85.83% and 78.86% specificity, along with 94.09% and 41.56% sensitivity, for disease diagnosis and lung sound classification, respectively.).

This study introduces an augmented reality (AR) system using electromagnetic (EM) stylet tracking to enhance surgical precision and patient safety for external ventricular drainage (EVD). By integrating a tablet PC, 3D camera, and EM tracking system, this study developed a real-time navigation method for catheter insertion. The system employs deep learning facial landmark detection and coordinate registration techniques to align CT, 3D camera, and EM tracking systems. A phantom model experiment was conducted with 20 trials: 20 drainage procedures and 20 accuracy measurements. All 20 drainage attempts were successful. And the Euclidean distance error for the 20 trials of measurement was 1.68 ± 0.66 mm. The research suggests significant potential for further development of AR technologies in neurosurgical interventions, highlighting the importance of continuous technological innovation in medical precision procedures.