Biomedical Engineering Letters最新文献

Parkinson's disease is a neurodegenerative disorder principally manifested as motor disabilities. In clinical practice, diagnostic rating scales are available for broadly measuring, classifying, and characterizing the disease progression. Nonetheless, these scales depend on the specialist's expertise, introducing a high degree of subjectivity. Thus, diagnosis and motor stage identification may be affected by misinterpretation, leading to incorrect or misguided treatments. This work addresses how to learn multimodal representations based on compact gait and eye motion descriptors whose fusion improves disease diagnosis prediction. This work introduces a noninvasive multimodal strategy that combines gait and ocular pursuit motion modalities into a geometrical Riemannian Neural Network for PD quantification and diagnostic support. Markerless gait and ocular pursuit videos were first recorded as Parkinson's observations, which are represented at each frame by a set of frame convolutional deep features. Then, Riemannian means are computed per modality using frame-level covariances coded from convolutional deep features. Thus, a geometrical learning representation is adjusted by Riemannian means, following early, intermediate, and late fusion alternatives. The adjusted Riemannian manifold combines input modalities to obtain PD prediction. The geometrical multimodal approach was validated in a study involving 13 control subjects and 19 PD patients, achieving a mean accuracy of 96% for early and intermediate fusion and 92% for late fusion, increasing the unimodal accuracy results obtained in the gait and eye movement modalities by 6 and 8%, respectively. The proposed method was able to discriminate Parkinson's patients from healthy subjects using multimodal geometrical configurations based on covariances descriptors. The covariance representation of video descriptors is highly compact (with an input size of 625 and an output size of 256 (1 BiRe)), facilitating efficient learning with a small number of samples, a crucial aspect in medical applications.

The limited imaging depth of optical endoscope restrains the identification of tissues under surface during the minimally invasive spine surgery (MISS), thus increasing the risk of critical tissue damage. This study is proposed to improve the accuracy and effectiveness of automatic spinal soft tissue identification using a forward-oriented ultrasound endoscopic system. Total 758 ex-vivo soft tissue samples were collected from ovine spines to create a dataset with four categories including spinal cord, nucleus pulposus, adipose tissue, and nerve root. Three conventional methods including Gray-level co-occurrence matrix (GLCM), Empirical Wavelet Transform (EWT), Variational Mode Decomposition (VMD) and two deep-learning based methods including Densely Connected Neural Network (DenseNet) model, one-dimensional Vision Transformer (ViT) model, were applied to identify the spinal tissues. The two deep learning methods outperformed the conventional methods with both accuracy over 95%. Especially the signal-based method (ViT) achieved an accuracy of 98.31% and a specificity of 99.2%, and the inference latency was only 0.0025 s. It illustrated the feasibility of applying the forward-oriented ultrasound endoscopic system for real-time intraoperative recognition of critical spinal tissues to enhance the precision and safety of minimally invasive spine surgery.

Breast cancer (BC) remains a significant global health issue, necessitating innovative methodologies to improve early detection and diagnosis. Despite the existence of intelligent deep learning models, their efficacy is often limited due to the oversight of small-sized masses, leading to false positive and false negative outcomes. This research introduces a novel segmentation-guided classification model developed to increase BC detection accuracy. The designed model unfolds in two critical phases, each contributing to a comprehensive BC diagnostic pipeline. In Phase I, the Attention U-Net model is utilized for BC segmentation. The encoder extracts hierarchical features, while the decoder, supported by attention mechanisms, refines the segmentation, focusing on suspicious regions. In Phase II, a novel ensemble approach is introduced for BC classification, involving various feature extraction methods, base classifiers, and a meta-classifier. An ensemble of model classifiers-including support vector machine, decision trees, k-nearest neighbor and artificial neural network- captures diverse patterns within these features. The Random Forest meta-classifier amalgamates their outputs, leveraging their collective strengths. The proposed integrated model accurately identifies different breast tumor classes, including malignant, benign, and normal. The precise region-of-interest analysis from segmentation phase significantly boosted classification performance of ensemble meta-classifier. The model accomplished an overall accuracy rate of 99.57% with high segmentation performance of 95% f1-score, illustrating its high discriminative power in detecting malignant, benign, and normal cases within the ultrasound image dataset. This research contributes to reducing breast tumor morbidity and mortality by facilitating early detection and timely intervention, ultimately supporting better patient outcomes.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-024-00435-7.

With the advent of robot-assisted surgery, user-friendly technologies have been applied to the da Vinci surgical system (dVSS), and their efficacy has been validated in worldwide surgical fields. However, further improvements are required to the traditional manipulation methods, which cannot control an endoscope and surgical instruments simultaneously. This study proposes a speech recognition control interface (SRCI) for controlling the endoscope via speech commands while manipulating surgical instruments to replace the traditional method. The usability-focused comparisons of the newly proposed SRCI-based and the traditional manipulation method were conducted based on ISO 9241-11. 20 surgeons and 18 novices evaluated both manipulation methods through the line tracking task (LTT) and sea spike pod task (SSPT). After the tasks, they responded to the globally reliable questionnaires: after-scenario questionnaire (ASQ), system usability scale (SUS), and NASA task load index (TLX). The completion times in the LTT and SSPT using the proposed method were 44.72% and 26.59% respectively less than the traditional method, which shows statistically significant differences (p < 0.001). The overall results of ASQ, SUS, and NASA TLX were positive for the proposed method, especially substantial reductions in the workloads such as physical demands and efforts (p < 0.05). The proposed speech-mediated method can be a candidate suitable for the simultaneous manipulation of an endoscope and surgical instruments in dVSS-used robotic surgery. Therefore, it can replace the traditional method when controlling the endoscope while manipulating the surgical instruments, which contributes to enabling the continuous surgical flow in operations consequentially.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-024-00429-5.

Osteoarthritis (OA) is a musculoskeletal disorder that affects weight-bearing joints like the hip, knee, spine, feet, and fingers. It is a chronic disorder that causes joint stiffness and leads to functional impairment. Knee osteoarthritis (KOA) is a degenerative knee joint disease that is a significant disability for over 60 years old, with the most prevalent symptom of knee pain. Radiography is the gold standard for the evaluation of KOA. These radiographs are evaluated using different classification systems. Kellgren and Lawrence's (KL) classification system is used to classify X-rays into five classes (Normal = 0 to Severe = 4) based on osteoarthritis severity levels. In recent years, with the advent of artificial intelligence, machine learning, and deep learning, more emphasis has been given to automated medical diagnostic systems or decision support systems. Computer-aided diagnosis is needed for the improvement of health-related information systems. This survey aims to review the latest advances in automated radiographic classification and detection of KOA using the KL system. A total of 85 articles are reviewed as original research or survey articles. This survey will benefit researchers, practitioners, and medical experts interested in X-rays-based KOA diagnosis and prediction.

Patients suffering from various neurological disorders, including major depressive disorder (MDD), often exhibit abnormal brain connectivity. In particular, patients with MDD show atypical brain oscillations propagation. This study aims to investigate an association between abnormal brain connectivity and atypical oscillatory propagation of electroencephalogram (EEG) signals in patients with a history of MDD. Previous findings of functional hyperconnectivity in beta oscillations (15-25 Hz) lead us to hypothesize that patients would experience abnormal beta oscillation propagation. Using the local phase gradient (LPG) method, we analyze a publicly available EEG dataset recorded during a probabilistic learning task. Our findings indicate that, upon receiving positive feedback during the learning task, patients with a history of MDD show more pronounced propagation directions of beta oscillations observed in the right frontal region compared to healthy controls. This directional pattern may help differentiate patients with a history of MDD from healthy controls. The observed abnormalities in brain oscillation propagation suggest that cognitive deficits in patients with a history of MDD might stem from excessive and negatively biased information transmission between brain regions.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-024-00433-9.

This study aims to create a fatigue recognition system that utilizes electroencephalogram (EEG) signals to assess a driver's physiological and mental state, with the goal of minimizing the risk of road accidents by detecting driver fatigue regardless of physical cues or vehicle attributes. A fatigue state recognition system was developed using transfer learning applied to partial ensemble averaged EEG power spectral density (PSD). The study utilized layer-wise relevance propagation (LRP) analysis to identify critical cortical regions and frequency bands for effective fatigue discrimination. A total of 21 participants were included in the study, and data augmentation techniques were used to enhance the system's classification accuracy. The results indicate a significant improvement in classification accuracy, particularly with the application of data augmentation. The classification accuracies were 99.2 ± 2.3% for the training data, 97.9 ± 3.1% for the validation data, and 96.9 ± 3.3% for the test data. This study advances the development of personalized EEG-based fatigue monitoring systems that have the potential to improve road safety and reduce accidents. The findings highlight the utility of EEG signals in detecting fatigue and the benefits of data augmentation in improving system performance. Further research is recommended to optimize data augmentation strategies and enhance the scalability and efficiency of the system.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-024-00431-x.

Accurate assessment of shoulder range of motion (ROM) is crucial for evaluating patient progress. Traditional manual goniometry often lacks precision and is subject to inter-observer variability, especially in measuring shoulder internal rotation (IR). This study introduces an artificial intelligence (AI)-based approach that uses clinical photography to improve the accuracy of ROM quantification. We analyzed a total of 150 clinical photographs, including 100 shoulder and 50 elbow images, taken between January and April 2022. An MMPose model with an HR-NET backbone architecture, pre-trained on the COCO-WholeBody dataset, was used to detect 17 anatomical landmarks. A random forest classifier (PoseRF) then categorized poses, and ROM angles were calculated. Concurrently, two clinicians independently measured shoulder IR at the vertebral level, and inter-observer agreement was evaluated. Linear regression analyses were conducted to correlate the AI-derived measurements with the clinicians' assessments. The AI-based algorithm accurately detected anatomical landmarks in 96% of shoulder and 100% of elbow images. Pose detection achieved 95% accuracy overall, with 100% accuracy for specific shoulder (abduction, flexion, external rotation) and elbow (flexion, extension) poses. Intraclass correlation coefficients (ICCs) between the AI algorithm and human observers ranged from 0.965 to 0.997, indicating excellent inter-observer reliability. Kruskal-Wallis test showed no statistically significant differences in ROM measurements among the AI algorithm and two human observers across all joint angles (p > 0.05). The AI-based algorithm demonstrated performance comparable to that of human observers in quantifying shoulder and elbow ROM from clinical photographs. For shoulder internal rotation, the AI approach showed potential for improved consistency compared to traditional methods.

Supplementary information: The online version contains supplementary material available at 10.1007/s13534-024-00432-w.

MR-only radiotherapy planning is beneficial from the perspective of both time and safety since it uses synthetic CT for radiotherapy dose calculation instead of real CT scans. To elevate the accuracy of treatment planning and apply the results in practice, various methods have been adopted, among which deep learning models for image-to-image translation have shown good performance by retaining domain-invariant structures while changing domain-specific details. In this paper, we present an overview of diverse deep learning approaches to MR-to-CT synthesis, divided into four classes: convolutional neural networks, generative adversarial networks, transformer models, and diffusion models. By comparing each model and analyzing the general approaches applied to this task, the potential of these models and ways to improve the current methods can be can be evaluated.

Deep learning-based image registration methods offer advantages of time efficiency and registration outcomes by automatically extracting enough image features. Currently, more and more scholars choose to use cascaded networks to achieve coarse-to-fine registration. Although cascaded networks take a lot of time in the training and inference stages, they can improve registration performance. In this study, we utilize the advantage of high registration performance of cascaded networks. Two VoxelMorph convolutional neural networks are cascaded together. The first VoxelMorph network outputs the dense deformation field of registration. The second network outputs a noisy deformation field, which serves to boost the registration performance by minimizing the error in comparison with Gaussian noise. At the same time, the Enhancement Features-encoder (EF-encoder) block is introduced in the encoder and decoder part of the network to achieve enhancement features functions by attention mechanism. This paper conducted experiments on LPBA40 and HBN datasets. The experimental results show that the Dice similarity coefficient, Average Symmetric Surface Distance, Structural similarity and Pearson correlation coefficient of GaussianMorph are better than those of VM, VM × 2 and TST-Net. Experimental results show that GaussianMorph can improve the registration accuracy.