Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference最新文献

The rising global prevalence of heart disease necessitates early detection for improved diagnosis and treatments. Automated echocardiography robotic systems are revolutionizing cardiology by enhancing diagnostic accuracy and efficiency. These systems integrate real-time image acquisition and processing to navigate patient anatomy and adapt imaging techniques dynamically without human intervention. Accurate cardiac view classification is vital for capturing diagnostically relevant images, forming the basis for subsequent automated disease detection and diagnosis. Although deep learning has emerged as a powerful tool for medical image analysis, its application in echocardiography remains limited due to the complexity of multi-view echocardiography imaging. The proposed system leverages deep learning models, specifically convolutional neural networks, trained on a diverse dataset of echocardiographic images to distinguish standard cardiac views, including the parasternal long-axis, parasternal short-axis, and apical four-chamber views. This capability enables the robotic system to autonomously navigate patient anatomy and optimize image acquisition in real time, minimizing operator dependency and ensuring imaging consistency. The long-term objective of this study is to develop a fully autonomous robotic system capable of early and accurate cardiovascular disease diagnosis, ultimately reducing diagnostic delays and improving patient outcomes.

This study aims to investigate whether quantitative radiomic features extracted from Positron Emission Tomography/Computed Tomography (PET/CT) could differentiate triple-negative breast cancer (TNBC) from non-triple-negative breast cancer (non-TNBC). We propose a pipeline that combines deep learning for cancer lesion segmentation with machine learning techniques to classify TNBC. Our approach leveraged the radiomic features extracted from ¹⁸F-fluorodeoxyglucose PET/CT. This retrospective study included the PET/CT images of 217 patients with breast cancer (57 TNBC and 160 non-TNBC) admitted to Georges-François Leclerc Hospital. The tumor regions of interest were automatically segmented on PET images using a deep learning model and mapped to CT scans. Radiomic features were extracted from 3D tumor volumes and machine learning classifiers were built using stratified 5-fold cross-validation. Recursive feature elimination was employed to rank and select the most relevant radiomic features, thereby enhancing classification performance. The model was evaluated using the F1-score, area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, and specificity. The proposed method achieved promising performance, with an F1-score of 0.90 ± 0.02, an accuracy of 0.86 ± 0.07, a sensitivity of 0.91 ± 0.06, and an AUC of 0.88 ± 0.04, using the top-ranked features. The metrics were evaluated as the average over a five-fold cross-validation. Radiomic features extracted from PET and CT scans provide valuable prognostic insights for the identification of TNBC. This study demonstrated that machine learning algorithms based on radiomic features and automated PET/CT segmentation can accurately distinguish TNBC from non-TNBC.Clinical relevance- This study demonstrates the potential of image-based radiomic analysis combined with machine learning to differentiate triple-negative breast cancer (TNBC) from non-TNBC. By using deep learning for automatic tumor segmentation and feature extraction, this approach offers a non-invasive, quantitative tool that can improve TNBC diagnosis and the efficiency of treatment strategies. These advancements may help clinicians provide more reliable insights, while reducing the likelihood of misclassification.

For adolescent idiopathic scoliosis (AIS), a common condition in children, physiotherapy scoliosis-specific exercise (PSSE) is an effective conservative treatment. However, the long-term process of PSSE treatment often leads to low compliance during unsupervised exercises. In this study, we proposed a wearable system for the evaluation of PSSE compliance for AIS patients. The proposed system contains wearable devices and analysis software. The wearable device collected surface electromyography (sEMG) data from back muscles. We extracted features from sEMG data, and adopted support vector machine classifiers to evaluate PSSE compliance for AIS patients in the software. To validate the proposed system, we collected data from 11 AIS patients during a typical exercise in PSSE. Among the extracted features, the most promising for differentiating PSSE compliance were those related to electromyography (EMG) amplitude and muscle fatigue. Specifically, the integrated EMG and frequency ratio showed strong potential. To evaluate the proposed system, we adopted leave-one-subject-out cross-validation, resulting in perfect accuracy. The results showed that the proposed system was potentially feasible for evaluating PSSE compliance in AIS patients to achieve optimal efficacy, and was convenient for supporting clinicians and parents in monitoring correction of AIS patients' PSSE execution.Clinical Relevance- This system provides a method for evaluating PSSE compliance in AIS patients, helping achieve optimal PSSE efficacy.

Chest CT scans are essential in diagnosing lung abnormalities, including lung cancer, but their utility in training deep learning models is often pushed back by limited data availability, high labeling costs, and privacy concerns. To address these challenges, this study explores the use of score-based diffusion models for the conditional generation of lung CT scans slices. Two generation scenarios are explored: one limited to lung segmentation masks and another incorporating both lung and nodule segmentation mappings to guide the synthesis process. The proposed methods are custom U-Net architecture models trained to predict the scores in Variance Preserving (VP) and Variance Exploding (VE) Stochastic Differential Equations (SDEs), composing the primary ground for comparison in conditional sample generation. The results demonstrate the VP SDEs model's superiority in generating high-fidelity images, as evidenced by high SSIM (0.894) and PSNR (28.6) values, as well as low domain-specific FID (173.4), MMD (0.0133) and ECS (0.78) scores. The generated images consistently followed the conditional mapping guidance during the generation process, effectively producing realistic lung and nodule structures, highlighting their potential for data augmentation in medical imaging tasks. While the models achieved notable success in generating accurate 2D lung CT scan slices given simple conditional image region mappings, future work surrounds the extension of these methods to 3D conditional generation and the use of richer conditional mappings to account for broader anatomical variations. Nevertheless, this study holds promise for improvement in computer-aided systems through the support in deep learning model training for lung disease diagnosis and classification.

Early detection of burnout is of utmost importance to avoid severe health consequences. Burnout is typically assessed through standardized questionnaires with self-reported information, a technique that could potentially delay its diagnosis. Wearable devices continuously and unobtrusively collect health-related data, making them valuable tools for the early detection of several mental health issues, including burnout syndrome. In this paper we report initial insights on the machine learning prediction of baseline burnout risk across cognitive, emotional, and physical dimensions. Our data consists of the first 30 days of a 9-months longitudinal study with 239 participants, including monthly burnout assessments and health data from smartwatches. Aggregated mean and standard deviation of physiological features over time windows of varying duration were employed as predictors of baseline burnout risk. Models employing sleep, cardiac, and stress features achieved a balanced accuracy of 0.66 and 0.68 in the detection of cognitive weariness and physical fatigue risk, respectively. The prediction of emotional exhaustion risk reached lower performance with a balanced accuracy of 0.55, suggesting the need of integrating additional data sources to reach better-than-chance performance. We expect to improve burnout risk prediction by crafting additional features and exploiting the collected data over their full longitudinal scale.

Modeling the spatiotemporal patterns of whole-brain functional networks (FBNs) using functional magnetic resonance imaging (fMRI) is crucial for understanding brain function. Although existing methods, either shallow or deep models, have achieved promising outcomes, they lack the capability to concurrently extract multiple target FBNs while fully leveraging the inherent four-dimensional (4D) features of fMRI data. In this study, we propose a Multi-Pattern Spatiotemporal Hybrid Attention 4D CNN model (MSTHA-4DCNN) to concurrently capture the spatiotemporal patterns of multiple FBNs, building upon the rich spatial and temporal characteristics embedded in 4D fMRI data. The MSTHA-4DCNN extracts spatial patterns through the Multi-Pattern Spatial Attention 4D CNN (MSA-4DCNN), and subsequently incorporates Multi-Pattern Temporal Guided Attention Network (MT-GANet) to model temporal representations guided by the derived spatial patterns. We train the proposed model on a naturalistic fMRI dataset, and evaluate its generalizability on an independent public dataset from Cambridge Centre for Ageing and Neuroscience (Cam-CAN). The experimental results indicate that MSTHA-4DCNN exhibits promising performance and generalization ability in concurrently and effectively identifying spatiotemporal patterns of FBNs, outperforming other state-of-the-art models and offering a potent tool for advancing our understanding of complex neural processes.

Limb movement coordination is a critical indicator in general movement analysis (GMA), which is often used to assess newborn neurological development. Asymmetry in limb movements may indicate brain injury or motor control disorders, also associated with conditions such as cerebral palsy. In this work, we present an automated video processing framework for assessing the coordination of left and right limb movements, aiming to assist healthcare professionals to evaluate infant's limb movement coordination during GMA. We use AggPose, a pose recognition tool based on a Transformer architecture, to extract 12 keypoints (including arms and legs) from video frames. The intensity of movement is calculated using the temporal standard deviation of the keypoint coordinates. Finally, the coordination of movement is analyzed by comparing the cross-correlation and Pearson correlation coefficients of the movement signals between left and right limbs. Our clinical dataset, created in the neonatal intensive care unit, includes 23 preterm infants without neurological disorders. The proposed method shows average cross-correlation and Pearson correlation coefficients of 0.788 and 0.712, respectively, indicating the potential in analyzing the motion coordination of infant limb movements.

A Cochlear Implant (CI) is the only option to treat patients with profound hearing loss, but requires the tuning of a large number of parameters. However, the necessary objective and reproducible patient-specific data are still missing. Building on recent research using reaction times to acoustic spectrotemporal modulations, we developed a realistic simulator of the auditory system, considering the CI, the electrode output, the auditory nerve population, and the ultimate decision of stimulus detection by the brain. The simulation results are in line with measured Reaction-Time tests to spectrotemporal modulations, while also showing some resemblance to actual patient data.Clinical relevance- The simulator forms the basis for an optimisation method to find optimal CI parameters for each patient, necessitating only a limited number of measurements.

The brain exhibits multiple global states that can be distinguished by wave patterns in electroencephalogram (EEG). Global brain states during the wake have been of much interest among neurologists recently, but some of the states occur in low percentages, such as less than 10% of all epochs. This scarcity makes automatic classification difficult due to class imbalance. By introducing resampling and data augmentation techniques, we developed a system based on ResNet that automatically classifies highly imbalanced brain state data. Namely, we tested oversampling, undersampling, and SMOTE on single-channel EEG recordings obtained from mice. The results showed the effectiveness of dealing with class imbalance, opening possibilities for further analysis of global brain states during the wake.

Accurate ovulation prediction is crucial for fertility management. Although calendar-based methods are widely used, they are effective only for regular cycles. Furthermore, machine learning-based ovulation prediction studies obtained declined accuracy for irregular cycles and became less reliable. To address this limitation, we propose an ovulation prediction framework that (i) generates novel features by integrating temporal heart rate variability (HRV) patterns from ECG with temporal temperature values and 10-min resolution temperature features, and (ii) employs a light gradient boosting machine (LGBM). Participants wore an ECG device and a temperature sensor during sleep. The prediction focused on a 8-day period, covering 5 days before to 2 days after ovulation, to capture key physiological changes in the fertile window and improve prediction performance. The proposed framework obtained an area under the receiver operating characteristic curve (AUROC) of 0.73 and their performance was superior performance when compared with various machine and deep learning models. Notably, the model excelled in predicting ovulation for irregular cycles, achieving AUROC of 0.84 in the highly irregular group and 0.88 in the undefined group. These findings highlight the importance of temporal segmentation and multimodal feature integration for enhanced ovulation prediction. The proposed framework accurately predicts the ovulation date up to 5 days in advance for premenopausal women, significantly enhancing fertility management.