Medical & Biological Engineering & Computing最新文献

Three-dimensional (3D) ultrasound imaging offers a larger field of view and enables volumetric measurements. Among the versatile methods, free-hand 3D ultrasound imaging utilizing deep learning networks for spatial coordinate prediction exhibits advantages in terms of simplified device configuration and user-friendliness. However, this imaging method is restricted to predicting the relative spatial transformation between two consecutive 2D ultrasound images, resulting in substantial cumulative errors. When imaging large organs, cumulative errors can severely distort the 3D images. In this study, we proposed a labeling strategy based on the ultrasound image coordinate system, enhancing the network prediction accuracy. Meanwhile, pre-planning the scanning trajectory and using it to guide the network prediction significantly reduced cumulative error. Spinal 3D ultrasound imaging was performed on both healthy volunteers and scoliosis patients. Comparison of reconstruction results across different methods demonstrated that the proposed method improved the prediction accuracy by approximately 40% and reduced the cumulative error by nearly 80%. This method shows promise for application in various deep learning networks and different tissues and is expected to facilitate the broader clinical adoption of 3D ultrasound imaging.

Numerical simulations and clinical measurements of nasal resistance are in quantitative disagreement. The order of magnitude of this mismatch, that sometimes exceeds 100%, is such that known sources of uncertainty cannot explain it. The goal of the present work is to examine a source of bias introduced by the design of medical devices, which has not been considered until now as a possible explanation. We study the effect of the location of the probe on the rhinomanometer that is meant to measure the ambient pressure. Rhinomanometry is carried out on a 3D silicone model of a patient-specific anatomy; a clinical device and dedicated sensors are employed side-by-side for mutual validation. The same anatomy is also employed for numerical simulations, with approaches spanning a wide range of fidelity levels. We find that the intrinsic uncertainty of the numerical simulations is of minor importance. To the contrary, the position of the pressure tap intended to acquire the external pressure in the clinical device is crucial, and can cause a mismatch comparable to that generally observed between in-silico and in-vivo rhinomanometry data. A source of systematic bias may therefore exist in rhinomanometers, designed under the assumption that measurements of the nasal resistance are unaffected by the flow development within the instruments.

The validation of multimodal deep learning models for medical image translation is limited by the lack of high-quality, paired datasets. We propose a novel framework that leverages computational phantoms to generate realistic CT and MRI images, enabling reliable ground-truth datasets for robust validation of artificial intelligence (AI) methods that generate synthetic CT (sCT) from MRI, specifically for radiotherapy applications. Two CycleGANs (cycle-consistent generative adversarial networks) were trained to transfer the imaging style of real patients onto CT and MRI phantoms, producing synthetic data with realistic textures and continuous intensity distributions. These data were evaluated through paired assessments with original phantoms, unpaired comparisons with patient scans, and dosimetric analysis using patient-specific radiotherapy treatment plans. Additional external validation was performed on public CT datasets to assess the generalizability to unseen data. The resulting, paired CT/MRI phantoms were used to validate a GAN-based model for sCT generation from abdominal MRI in particle therapy, available in the literature. Results showed strong anatomical consistency with original phantoms, high histogram correlation with patient images (HistCC = 0.998 ± 0.001 for MRI, HistCC = 0.97 ± 0.04 for CT), and dosimetric accuracy comparable to real data. The novelty of this work lies in using generated phantoms as validation data for deep learning-based cross-modality synthesis techniques.

Sperm head morphology has been identified as a characteristic that can be used to predict a male's semen quality. Here, harnessing the close relationship considering sperm head shape to quality and morphology, we propose a joint learning model for sperm head segmentation and morphological category prediction. In the model, the sperm category prediction and the ellipticity, calculated by using the segmented sperm head profile, are used to synthesize the morphology to which the sperm belongs. In traditional clinical testing, fertility experts analyze sperm morphology by 2D images of sperm samples, which cannot represent the whole character of their quality and morphological category. To overcome the problem that single-angle 2D images cannot accurately identify sperm morphology, we use a deep-learning-based tracking and detection system to dynamically acquire sperm images with multiple frames and angles and then use the multi-frame and multi-angle time-series images of sperm to determine sperm morphology based on the multi-task model proposed in this study. Performing better than 3D sperm reconstruction and traditional computer-assisted sperm assessment systems, this approach enables end-to-end analysis of viable spermatozoa, requiring minimal computing power and utilizing equipment already available in most embryology laboratories.

Fetal standard plane detection is essential in prenatal care, enabling accurate assessment of fetal development and early identification of potential anomalies. Despite significant advancements in machine learning (ML) in this domain, its integration into clinical workflows remains limited-primarily due to the lack of standardized, end-to-end operational frameworks. To address this gap, we introduce FetalMLOps, the first comprehensive MLOps framework specifically designed for fetal ultrasound imaging. Our approach adopts a ten-step MLOps methodology that covers the entire ML lifecycle, with each phase meticulously adapted to clinical needs. From defining the clinical objective to curating and annotating fetal US datasets, every step ensures alignment with real-world medical practice. ETL (extract, transform, load) processes are developed to standardize, anonymize, and harmonize inputs, enhancing data quality. Model development prioritizes architectures that balance accuracy and efficiency, using clinically relevant evaluation metrics to guide selection. The best-performing model is deployed via a RESTful API, following MLOps best practices for continuous integration, delivery, and performance monitoring. Crucially, the framework embeds principles of explainability and environmental sustainability, promoting ethical, transparent, and responsible AI. By operationalizing ML models within a clinically meaningful pipeline, FetalMLOps bridges the gap between algorithmic innovation and real-world application, setting a precedent for trustworthy and scalable AI adoption in prenatal care.

Patients with lower limb impairments often face sit-to-stand-to-sit motion challenges. The patients utilize a greater trunk flexion angle at seat-off time to mitigate knee moment. Alternative methods of STSTS motion strategies are required to study and understand the various patterns to guide physical rehabilitation programs in clinical practice. Four different STSTS strategies-Natural, Full Flexion, Pelvis-spine alignment, and Frame-Assisted-were experimented with twenty healthy subjects in a 3D motion capture lab, and inverse kinematics and dynamics methods were used for motion analysis in Visual 3D. At seat-off time in full flexion, the maximum trunk flexion angle is 58.77(± 17.92) degrees, duration is 1.63 s, 27% of the cycle, which reduces knee moment by -0.466(± 0.2) N.m/kg, increased hip moment by 0.67(± 0.312) N.m/kg, and ankle moment by 0.225(± 0.09) N.m/kg for the compensation. The compensatory movement also occurred while sitting down. Frame-assisted STSTS motion reduced knee moments without increases in hip and ankle moments at the maximum of trunk flexion angle while standing and sitting, and its motion patterns are similar to pelvis-spine alignment and natural strategies. These findings provide valuable insights for physiotherapists to predict the current stage of the patient for clinical assessment and guide in the design and development of medical devices.

Although biomedical engineering (BME) is a profession with ethical responsibilities comparable to those in medicine, it has, until now, lacked a counterpart to the Hippocratic Oath. While professional societies have established codes of ethics for biomedical engineers, these documents lack the symbolic and ceremonial significance of an oath or pledge. By contrast, the recitation of the Hippocratic Oath, or its modern version, the "Physician's Pledge," serves as a powerful rite of passage for medical students, fostering a strong sense of ethical duty at the start of their professional journey. However, the content of the Hippocratic Oath includes elements specific to clinical practice and is not directly applicable to biomedical engineering. To fill this gap, we have created a "Biomedical Engineer's Pledge," comprising a preamble, ten promises, and a concluding statement, to inspire ethical awareness and establish a meaningful graduation tradition.

Minimally invasive neurosurgery presents specific challenges due to the limited operative space and complex cranial anatomy, requiring highly precise and safe surgical guidance. Augmented Reality (AR) technology offers the potential to improve surgical accuracy and safety by overlaying critical digital information onto real-world surgical environments. In this study, we present a study that aims to compare four AR visualization methods-2D flat display, smart tablet, head-mounted display (HMD), and 3D autostereoscopic display-in guiding minimally invasive neurosurgical procedures, specifically focusing on ventriculocentesis. The effectiveness of the AR methods was evaluated through comprehensive user studies involving 32 participants (including 11 experienced surgeons), with assessment focused on critical performance metrics including accuracy, completion time, usability, and cognitive workload during simulated surgical procedures. Results demonstrated that 3D visualization methods significantly outperformed traditional 2D approaches in terms of puncture accuracy and angular precision. Specifically, surgeons showed a statistically significant improvement in localization accuracy, with mean error reduced from 2.69 mm to 1.67 mm, and angular deviation from 5.62° to 1.54°. In comparing the two 3D visualization systems, the HMD exhibited superior task completion efficiency, while the 3D autostereoscopic display demonstrated higher usability scores and lower perceived workload ratings. Notably, the 3D systems effectively reduced the performance disparity between novice and experienced practitioners, suggesting their potential to accelerate the learning curve for less experienced users. We conclude that AR holds significant potential to enhance performance and decision-making in minimally invasive neurosurgical guidance.

Persistent atrial fibrillation is the most common sustained cardiac arrhythmia, frequently linked with increased mortality and morbidity. Electrical cardioversion (ECV) remains the gold standard for sinus rhythm (SR) restoration, even though presenting potential adverse effects and a high relapsing rate. Predicting ECV outcome from the 12-lead ECG could reduce healthcare costs while preventing complications in patients unlikely to maintain SR. To this end, atrial activity (AA) organization has been traditionally evaluated through the amplitude and dominant frequency of the fibrillatory waves at lead II. However, physiological systems are known to exhibit complex dynamics across multiple time-scales, making multiscale (MSE) entropy measures a more suitable tool, as they can incorporate relevant information that may have been previously overlooked. Here, the predictive power of different MSE-based indices for the ECV outcome in 58 patients is evaluated. AA was estimated using a QT segment cancellation algorithm. Patients were classified based on SR maintenance after a 30-day follow-up. Results show that traditionally used indices report the highest predictive rate over the limb leads (79%). However, they are outperformed by Refined MSE over precordial leads (87%). Moreover, when considering statistical modeling techniques such as support vector machines, the prediction accuracy is increased (98%). In conclusion, MSE-based indices computed from precordial leads can robustly predict ECV outcome with higher accuracy than traditional approaches.

Glioblastoma (GBM) carries poor prognosis; epidermal-growth-factor-receptor (EGFR) mutations further shorten survival. We propose a fully automated MRI-based decision-support system (DSS) that segments GBM and classifies EGFR status, reducing reliance on invasive biopsy. The segmentation module (UNet SI) fuses multiresolution, entropy-ranked shearlet features with CNN features, preserving fine detail through identity long-skip connections, to yield a Lightweight 1.9 M-parameter network. Tumour masks are fed to an Inception ResNet-v2 classifier via a 512-D bottleneck. The pipeline was five-fold cross-validated on 98 contrast-enhanced T1-weighted scans (Memorial Hospital; Ethics 24.12.2021/008) and externally validated on BraTS 2019. On the Memorial cohort UNet SI achieved Dice 0.873, Jaccard 0.853, SSIM 0.992, HD95 24.19 mm. EGFR classification reached Accuracy 0.960, Precision 1.000, Recall 0.871, AUC 0.94, surpassing published state-of-the-art results. Inference time is ≤ 0.18 s per slice on a 4 GB GPU. By combining shearlet-enhanced segmentation with streamlined classification, the DSS delivers superior EGFR prediction and is suitable for integration into routine clinical workflows.