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

Epilepsy, a prevalent neurological disease, often requires accurate identification of the seizure onset zone (SOZ) in the brain for successful surgical removal of this region. SOZ identification is a lengthy process that traditionally relies on the knowledge and experience of the neurosurgeon. Advancements in artificial intelligence have opened the door to automatic SOZ localization. This study proposes the application of autoencoder-based anomaly detection in SOZ electrode classification for the first time. We trained the autoencoder in a leave-one-patient-out manner with electrocorticography (ECoG) signals from intact channels. The anomaly feature was determined by the maximum error between original and reconstructed signals for each channel of the test patient. We investigated the usefulness of anomaly features along with the interictal epileptiform discharge (IED) biomarker feature. A linear support vector machine classifier achieved 70.49% accuracy with the anomaly feature, 64.71% accuracy with IED feature, and 74.44% accuracy with combined anomaly and IED features. The study demonstrates the effectiveness of anomaly detection in the direct localization of SOZs and suggests that multiple biomarker features can enhance automatic SOZ localization performance.Clinical Relevance- This study demonstrates the clinical relevance of using anomaly features from ECoG data for efficient SOZ localization. It highlights the effectiveness of combining anomaly features with IED biomarkers to enhance the automatic SOZ classification performance and improve surgical planning in epilepsy treatment.

Vagus Nerve Stimulation (VNS) is an effective therapy for drug-resistant epilepsy (DRE), but its efficacy varies across individuals. This study uses the weighted Phase Lag Index (wPLI) across multiple frequency bands to examine the dose-dependent regional brain synchronization effect at an individual level. EEG data from four DRE patients (two responders and two non-responders) were analyzed under controlled VNS intensities. Results show that non-responders exhibit an increased synchrony at higher intensity levels, particularly in the beta band and the broadband. Findings highlight individualized neural responses to VNS, underscoring the possibility for personalized stimulation strategies to optimize therapy.Clinical Relevance- Refining the VNS titration process through dose-dependent brain connectivity analysis could accelerate optimization, improving therapeutic outcomes and personalization for drug-resistant epilepsy patients.

This manuscript introduces a deep learning algorithm designed for spatial and temporal source reconstruction based on signals captured by MEG devices. Estimating brain signals at the source level is a significant challenge in magnetoencephalographic (MEG) data processing. Traditional algorithms excel in temporal resolution but face limitations in spatial resolution due to the inherently ill-posed nature of the problem. However, precise localization of pathological tissues is often crucial for providing reliable information to clinicians, which makes this a key area for improvement. Deep learning solutions have emerged as promising candidates for high-resolution signal estimation in this context. The proposed approach, called 'Deep-MEG', utilizes a hybrid neural network architecture capable of extracting both temporal and spatial information from MEG sensor signals. Unlike traditional methods, the algorithm can handle the entire brain, making it suitable for imaging not just cortical sources but also subcortical ones. To validate its performance, the authors conducted simulations with multiple active sources using a realistic forward model and compared the results with those from various state-of-the-art reconstruction algorithms.Clinical relevance- This study represent a first approach for accurate deep source localization and reconstruction leading to diagnosis support to clinicians.

Colon cancer is one of the deadliest types of cancer in the United States, with close to 50,000 projected deaths in 2024. The disease requires early diagnosis to optimize chances of survival by enabling timely administration of treatment. To investigate the key non-genetic (NG) factors influencing the onset of colon cancer and evaluate how genetic factors enhance the performance of machine learning (ML) models in predicting incidence, we incorporated polygenic risk scores (PRSs) alongside NG data in ML models to predict 10-year incident risk prediction of colon cancer using data from the UK Biobank. This approach enabled us to assess the added predictive value of PRSs in multi-modal models in estimating the 10-year risk of developing colon cancer over NG data alone. Moreover, our research focused on identifying the most relevant and predictive PRS and validating them using a robust ML framework. To ensure the robustness, we restricted the cohort to White British individuals to minimize ancestry-related heterogeneity. PRSs have proven effective in enhancing disease prediction for conditions such as breast cancer, myocardial infarction, and schizophrenia, reinforcing their relevance in clinical research. Exploring six PRSs, our goal was to minimize false negatives while simultaneously maximizing area under the receiver-operating characteristic curve (AUC), in order to improve early detection rates by identifying those who are at risk for colon cancer. This research shows that PRSs can be used to enhance overall predictive ability of ML models in colon cancer research over NG factors alone, bolstering the argument for incorporating PRSs into routine clinical practice. PRSs can also help minimize false negatives, a key feature for disease prediction models, as missed potential diagnoses are life-threatening.

Anterior temporal lobe resection (ATLR) results in seizure freedom in half of individuals with drug-resistant temporal lobe epilepsy (TLE). Some investigators have suggested that larger resections lead to greater chance of seizure freedom, while others report no relationship. In this study, we examine the relationship between resection size and seizure freedom through (i) total volume analysis and (ii) a mass univariate regional approach.Patient demographics and resection volumes were collected for 283 patients who underwent subsequent ATLR, and seizure freedom was measured after 12 months. Additionally, the percentage resection of each Desikan-Kiliany parcellated region was calculated. We computed the AUC to measure effect sizes and used Wilcoxon ranksum tests to assess significance.Total resection volumes were larger in males than females, and larger in right than left ATLR. However, when scaled to percentage of brain tissue resected, only the hemisphere difference remained. There was no significant association of total or regional resection volume with post-operative seizure freedom.Larger resections in males are due to their larger total brain volumes. Smaller left-sided resections reflect the more conservative surgical approach in the language dominant hemisphere. Within the normal ranges of a typical ATLR, larger resection volumes do not increase chance of seizure-freedom. Future studies should investigate the details of the resection of gray matter, such as piriform cortex, and white matter tracts that can form epileptogenic networks.

Acquisition of data on respiratory motion and anatomical deformation at the individual level is essential for improving the accuracy of surgical procedures and radiotherapy. Although time-series imaging using three-dimensional (3D) computed tomography (CT) and deep learning-based image interpolation have been explored, the acquisition of multiple imaging volumes remains a significant burden on patients because of breath-holding requirements and additional radiation exposure. In this study, we propose a framework for four-dimensional (4D) CT image generation, with images being generated at different time phases from a single-phase 3D chest CT image using only the magnitude of displacement. To achieve this, we employ a conditional diffusion model to generate displacement vector fields (DVFs) and propose a model that incorporates the initial-phase CT image and the mean DVF of the target phase as guidance. We trained and tested the proposed model using 4D-CT images from 62 cases and evaluated its effectiveness by deforming the 3D-CT image at the end-expiration phase using the predicted DVF. The validity of our approach was confirmed through quantitative comparisons under multiple guidance scenarios.Clinical Relevance- The proposed diffusion model predicts Deformation Vector Fields (DVFs) that capture respiratory motion, thereby enabling the generation of 3D-CT images at different respiratory phases from a single-phase CT scan. This approach could be directly applied to radiotherapy planning and we expect that it could improve radiation targeting accuracy.

The peri-implantation period of pregnancy in mammals is a critical stage for successful embryo implantation and offspring health. Significant molecular and biophysical interactions occur between the mother (uterine endometrium) and the embryo during this time. However, the underlying biological mechanisms contributing to pregnancy success, loss, or repercussions for offspring health remain unclear. Organ-ona-chip technology facilitates in vitro models of physiological cell and tissue environments in 3D, accurately replicating the behaviour of cells and tissues with appropriate flow conditions as observed in vivo. This study presents microfluidic devices designed to create an endometrium-on-a-chip to investigate mechanisms involved in successful pregnancy during the peri-implantation period. We assessed the channels' appropriate flow rate conditions, the devices' suitability for cell culture, and a suitable culture medium to maintain different endometrial cell types together in the devices. An in vivo culture model that recapitulates the complexities of the endometrial tissue will support effective studies and enhance our understanding of the mechanisms underpinning endometrial function in the peri-implantation period.Clinical Relevance- This allows for an in vitro culture model that mimics the biology of the endometrium, enabling high-throughput testing of mechanisms to deepen our understanding of reproductive biology.

Cardiac output (CO) is one of the most significant physiological parameters for cardiovascular monitoring. The gold standard for CO estimation, thermodilution, requires invasive catheterization, limiting its frequent use. Non-invasive methods exist but often lack accuracy. In this study, we applied deep learning to estimate CO based on different input signals including arterial pressure (ART), electrocardiography (ECG), photoplethysmography (PPG), and their combinations. Using publicly available VitalDB database, models were trained and evaluated according to different signals. Performance was measured using mean absolute error (MAE), root mean square error (RMSE), bias, and limits of agreement (LOA). Of all combinations tested, the triadic model of ART, ECG, and PPG yielded the best performance in MAE (0.66 L/min) and stronger correlation (R = 0.84) with reference CO values. The present study indicates the promise of deep learning for accurate noninvasive estimation of CO. Future research should emphasize the interpretability of the model, scaling up datasets, and facilitating real time applications for increased clinical utility.

Periodontitis is an inflammatory disease of the root surface caused by bacterial biofilms, leading to tooth loss and systemic health complications. Current treatment options rely on mechanical biofilm removal and chemical antimicrobial cleansing, which can be ineffective and contribute to tissue damage and bacterial resistance. Cold atmospheric plasma jet is a promising alternative due to its antimicrobial effects, lack of resistance development, and without known side effects. This study investigates the development and efficacy of a bipolar plasma nozzle designed for use on non-conductive surfaces, such as natural teeth. A proof-of-concept experiment was performed using a prototype plasma system (helium + 1 % oxygen used as working gas), with in vitro tests on bacterial agar plates of E. coli and S. aureus. The results demonstrate that plasma treatment effectively reduces bacterial concentration, with inhibition zones increasing with treatment duration, reaching two orders of magnitude reduction at 10 minutes of treatment. These findings support the potential of plasma technology as a novel method for periodontal disinfection, making way for further clinical applications.Clinical Relevance- This work provides proof-of-concept for the disinfection and treatment of inflamed surfaces in periodontology using a novel cold atmospheric plasma jet device.

Federated Learning (FL) provides a decentralized machine learning approach, where multiple devices or servers collaboratively train a model without sharing their raw data, thus enabling data privacy. This approach has gained significant interest in academia and industry due to its privacy-preserving properties, which are particularly valuable in the medical domain where data availability is often protected under strict regulations. A relatively unexplored area is applying FL to fine-tune Foundation Models (FMs) for time series forecasting, enhancing efficacy while preserving privacy. In this paper, we fine-tuned time series FMs with Electrocardiogram (ECG) and Impedance Cardiography (ICG) data using different FL techniques. We then examined various scenarios and discussed the challenges FL faces under different data heterogeneity configurations. Our empirical results demonstrated that while FL can be effective for fine-tuning FMs on time series forecasting tasks, its benefits depend on the data distribution across clients. We highlighted the trade-offs in applying FL to FM fine-tuning.