Search Results (1,189)

Background: Lung cancer ranks as the leading cause of cancer-related mortality worldwide. The complexity of tumor delineation, crucial for radiation therapy, requires expertise often unavailable in resource-limited settings. Artificial Intelligence (AI), particularly with advancements in deep learning (DL) and natural language processing (NLP), offers potential solutions yet is challenged by high false positive rates. Purpose: The Oncology Contouring Copilot (OCC) system is developed to leverage oncologist expertise for precise tumor contouring using textual descriptions, aiming to increase the efficiency of oncological workflows by combining the strengths of AI with human oversight. Methods: Our OCC system initially identifies nodule candidates from CT scans. Employing Language Vision Models (LVMs) like GPT-4V, OCC then effectively reduces false positives with clinical descriptive texts, merging textual and visual data to automate tumor delineation, designed to elevate the quality of oncology care by incorporating knowledge from experienced domain experts. Results: The deployment of the OCC system resulted in a 35.0% reduction in the false discovery rate, a 72.4% decrease in false positives per scan, and an F1-score of 0.652 across our dataset for unbiased evaluation. Conclusions: OCC represents a significant advance in oncology care, particularly through the use of the latest LVMs, improving contouring results by (1) streamlining oncology treatment workflows by optimizing tumor delineation and reducing manual processes; (2) offering a scalable and intuitive framework to reduce false positives in radiotherapy planning using LVMs; (3) introducing novel medical language vision prompt techniques to minimize LVM hallucinations with ablation study; and (4) conducting a comparative analysis of LVMs, highlighting their potential in addressing medical language vision challenges. Full article

Hemorrhage is the leading cause of preventable death in trauma care, requiring rapid and accurate detection to guide effective interventions. Hemorrhagic shock can be masked by underlying compensatory mechanisms, which may lead to delayed decision-making that can compromise casualty care. In this proof-of-concept study, we aimed to develop and evaluate machine learning models to predict Percent Estimated Blood Loss from a photoplethysmography waveform, offering non-invasive, field deployable solutions. Different model types were tuned and optimized using data captured during a hemorrhage and resuscitation swine study. Through this optimization process, we evaluated different time-lengths of prediction windows, machine learning model architectures, and data normalization approaches. Models were successful at predicting Percent Estimated Blood Loss in blind swine subjects with coefficient of determination values exceeding 0.8. This provides evidence that Percent Estimated Blood Loss can be accurately derived from non-invasive signals, improving its utility for trauma care and casualty triage in the pre-hospital and emergency medicine environment. Full article

The electroencephalogram (EEG), widely used for measuring the brain’s electrophysiological activity, has been extensively applied in the automatic detection of epileptic seizures. However, several challenges remain unaddressed in prior studies on automated seizure detection: (1) Methods based on CNN and LSTM assume that EEG signals follow a Euclidean structure; (2) Algorithms leveraging graph convolutional networks rely on adjacency matrices constructed with fixed edge weights or predefined connection rules. To address these limitations, we propose a novel algorithm: Dynamic Graph Convolutional Network with Dilated Convolution (DGDCN). By leveraging a spatiotemporal attention mechanism, the proposed model dynamically constructs a task-specific adjacency matrix, which guides the graph convolutional network (GCN) in capturing localized spatial and temporal dependencies among adjacent nodes. Furthermore, a dilated convolutional module is incorporated to expand the receptive field, thereby enabling the model to capture long-range temporal dependencies more effectively. The proposed seizure detection system is evaluated on the TUSZ dataset, achieving AUC values of 88.7% and 90.4% on 12-s and 60-s segments, respectively, demonstrating competitive performance compared to current state-of-the-art methods. Full article

Single-cell RNA sequencing (scRNA-seq) has revolutionized molecular biology and genomics by enabling the profiling of individual cell types, providing insights into cellular heterogeneity. Deep learning methods have become popular in single cell analysis for tasks such as dimension reduction, cell clustering, and data imputation. In this work, we introduce DropDAE, a denoising autoencoder (DAE) model enhanced with contrastive learning, to specifically address the dropout events in scRNA-seq data, where certain genes show very low or even zero expression levels due to technical limitations. DropDAE uses the architecture of a denoising autoencoder to recover the underlying data patterns while leveraging contrastive learning to enhance group separation. Our extensive evaluations across multiple simulation settings based on synthetic data and a real-world dataset demonstrate that DropDAE not only reconstructs data effectively but also further improves clustering performance, outperforming existing methods in terms of accuracy and robustness. Full article

Timely and accurate diagnosis of gastrointestinal diseases (GIDs) remains a critical bottleneck in clinical endoscopy, particularly due to the limited contrast and sensitivity of conventional white light imaging (WLI) in detecting early-stage mucosal abnormalities. To overcome this, this research presents Spectrum Aided Vision Enhancer (SAVE), an innovative, software-driven framework that transforms standard WLI into high-fidelity hyperspectral imaging (HSI) and simulated narrow-band imaging (NBI) without any hardware modification. SAVE leverages advanced spectral reconstruction techniques, including Macbeth Color Checker-based calibration, principal component analysis (PCA), and multivariate polynomial regression, achieving a root mean square error (RMSE) of 0.056 and structural similarity index (SSIM) exceeding 90%. Trained and validated on the Kvasir v2 dataset (n = 6490) using deep learning models like ResNet-50, ResNet-101, EfficientNet-B2, both EfficientNet-B5 and EfficientNetV2-B0 were used to assess diagnostic performance across six key GI conditions. Results demonstrated that SAVE enhanced imagery and consistently outperformed raw WLI across precision, recall, and F1-score metrics, with EfficientNet-B2 and EfficientNetV2-B0 achieving the highest classification accuracy. Notably, this performance gain was achieved without the need for specialized imaging hardware. These findings highlight SAVE as a transformative solution for augmenting GI diagnostics, with the potential to significantly improve early detection, streamline clinical workflows, and broaden access to advanced imaging especially in resource constrained settings. Full article

In recent years, data augmentation techniques have become the predominant approach for addressing highly imbalanced classification problems in machine learning. Algorithms such as the Synthetic Minority Over-sampling Technique (SMOTE) and Conditional Tabular Generative Adversarial Network (CTGAN) have proven effective in synthesizing minority class samples. However, these methods often introduce distributional bias and noise, potentially leading to model overfitting, reduced predictive performance, increased computational costs, and elevated cybersecurity risks. To overcome these limitations, we propose a novel architecture, FADEL, which integrates feature-type awareness with a supervised discretization strategy. FADEL introduces a unique feature augmentation ensemble framework that preserves the original data distribution by concurrently processing continuous and discretized features. It dynamically routes these feature sets to their most compatible base models, thereby improving minority class recognition without the need for data-level balancing or augmentation techniques. Experimental results demonstrate that FADEL, solely leveraging feature augmentation without any data augmentation, achieves a recall of 90.8% and a G-mean of 94.5% on the internal test set from Kaohsiung Chang Gung Memorial Hospital in Taiwan. On the external validation set from Kaohsiung Medical University Chung-Ho Memorial Hospital, it maintains a recall of 91.9% and a G-mean of 86.7%. These results outperform conventional ensemble methods trained on CTGAN-balanced datasets, confirming the superior stability, computational efficiency, and cross-institutional generalizability of the FADEL architecture. Altogether, FADEL uses feature augmentation to offer a robust and practical solution to extreme class imbalance, outperforming mainstream data augmentation-based approaches. Full article

Brain–computer interface (BCI) technology holds promise for improving motor rehabilitation in stroke patients. This review explores the immediate and long-term effects of BCI training, shedding light on the potential benefits and challenges. Clinical studies have demonstrated that BCIs yield significant immediate improvements in motor functions following stroke. Patients can engage in BCI training safely, making it a viable option for rehabilitation. Evidence from single-group studies consistently supports the effectiveness of BCIs in enhancing patients’ performance. Despite these promising findings, the evidence regarding long-term effects remains less robust. Further studies are needed to determine whether BCI-induced changes are permanent or only last for short durations. While evaluating the outcomes of BCI, one must consider that different BCI training protocols may influence functional recovery. The characteristics of some of the paradigms that we discuss are motor imagery-based BCIs, movement-attempt-based BCIs, and brain-rhythm-based BCIs. Finally, we examine studies suggesting that integrating BCIs with other devices, such as those used for functional electrical stimulation, has the potential to enhance recovery outcomes. We conclude that, while BCIs offer immediate benefits for stroke rehabilitation, addressing long-term effects and optimizing clinical implementation remain critical areas for further investigation. Full article

Alzheimer’s Disease and Dementia (ADD) progresses along a continuum of cognitive decline, typically from Subjective Cognitive Impairment (SCI) to Mild Cognitive Impairment (MCI) and eventually to dementia. While many studies have focused on classifying these clinical stages, fewer have examined whether brain connectomes encode this continuum in a low-dimensional, interpretable form. Motivated by the hypothesis that structural brain connectomes undergo complex yet compact changes across cognitive decline, we propose a Graph Neural Network (GNN)-based framework that embeds these connectomes into a two-dimensional manifold to capture the evolving patterns of structural connectivity associated with cognitive deterioration. Using attention-based graph aggregation and Principal Component Analysis (PCA), we find that MCI subjects consistently occupy an intermediate position between SCI and ADD, and that the observed transitions align with known clinical biomarkers of ADD pathology. This hypothesis-driven analysis is further supported by the model’s robust separation performance, with ROC-AUC scores of 0.93 for ADD vs. SCI and 0.81 for ADD vs. MCI. These findings offer an interpretable and neurologically grounded representation of dementia progression, emphasizing structural connectome alterations as potential markers of cognitive decline. Full article

The integration of vision and language processing into a cohesive system has already shown promise with the application of large language models (LLMs) in medical image analysis. Their capabilities encompass the generation of medical reports, disease classification, visual question answering, and segmentation, providing yet another approach to interpreting multimodal data. This survey aims to compile all known applications of LLMs in the medical image analysis field, spotlighting their promises alongside critical challenges and future avenues. We introduce the concept of X-stage tuning which serves as a framework for LLMs fine-tuning across multiple stages: zero stage, one stage, and multi-stage, wherein each stage corresponds to task complexity and available data. The survey describes issues like sparsity of data, hallucination in outputs, privacy issues, and the requirement for dynamic knowledge updating. Alongside these, we cover prospective features including integration of LLMs with decision support systems, multimodal learning, and federated learning for privacy-preserving model training. The goal of this work is to provide structured guidance to the targeted audience, demystifying the prospects of LLMs in medical image analysis. Full article

This research aims to offer a deep learning-based diagnostic approach for hemorrhagic complications linked to patent ductus arteriosus (PDA) in preterm infants. Utilizing the You Only Look Once (YOLO) algorithm, this research analyzed five key cardiac parameters derived from echocardiographic ultrasonic waves: the left ventricular ejection time (LVET), left ventricular internal dimension at diastole (LVIDd), left ventricular internal dimension at systole (LVIDs), posterior wall thickness at end-systole (HES), and RR interval between two successive R-waves. The proposed ensemble model achieved best-in-class detection accuracies for each parameter, with rates of 97.56% (LVET), 88.69% (LVIDd), 99.50% (LVIDs), 82.29% (HES), and 81.15% (RR interval). Furthermore, assessment of cardiac function using derived indices—end-systolic wall stress (ESWS) and rate-corrected mean velocity of circumferential fiber shortening (mVcfc)—achieved mean accuracy rates of 82.33% and 90.16%, respectively. This approach enables physicians to accurately evaluate cardiac function in preterm infants and facilitates the diagnosis of PDA-related hemorrhagic complications. Full article

Early prediction of lymph node metastasis (LNM) following neoadjuvant therapy (NAT) is crucial for timely treatment optimization in esophageal squamous cell carcinoma (ESCC). This study developed and validated a computed tomography-based radiomic model for predicting pathologically confirmed LNM status at the time of surgery in ESCC patients after NAT. A total of 469 ESCC patients from Sun Yat-sen University Cancer Center were retrospectively enrolled and randomized into a training cohort (n = 328) and a test cohort (n = 141). Three signatures were constructed: the tumor-habitat-based signature (Habitat_Rad), derived from radiomic features of three tumor subregions identified via K-means clustering; the multiple instance learning-based signature (MIL_Rad), combining features from 2.5D deep learning models; and the clinicoradiological signature (Clinic), developed through multivariate logistic regression. A combined radiomic nomogram integrating these signatures outperformed the individual models, achieving areas under the curve (AUCs) of 0.929 (95% CI, 0.901–0.957) and 0.852 (95% CI, 0.778–0.925) in the training and test cohorts, respectively. The decision curve analysis confirmed a high net clinical benefit, highlighting the nomogram’s potential for accurate LNM prediction after NAT and guiding individualized therapy. Full article

Accurate segmentation of cardiac MR images using deep neural networks is crucial for cardiac disease diagnosis and treatment planning, as it provides quantitative insights into heart anatomy and function. However, achieving high segmentation accuracy relies heavily on extensive, precisely annotated datasets, which are costly and time-consuming to obtain. This study addresses this challenge by proposing a novel data augmentation framework based on a condition-guided diffusion generative model, controlled by multiple cardiac labels. The framework aims to expand annotated cardiac MR datasets and significantly improve the performance of downstream cardiac segmentation tasks. The proposed generative data augmentation framework operates in two stages. First, a Label Diffusion Module is trained to unconditionally generate realistic multi-category spatial masks (encompassing regions such as the left ventricle, interventricular septum, and right ventricle) conforming to anatomical prior probabilities derived from noise. Second, cardiac MR images are generated conditioned on these semantic masks, ensuring a precise one-to-one mapping between synthetic labels and images through the integration of a spatially-adaptive normalization (SPADE) module for structural constraint during conditional model training. The effectiveness of this augmentation strategy is demonstrated using the U-Net model for segmentation on the enhanced 2D cardiac image dataset derived from the M&M Challenge. Results indicate that the proposed method effectively increases dataset sample numbers and significantly improves cardiac segmentation accuracy, achieving a 5% to 10% higher Dice Similarity Coefficient (DSC) compared to traditional data augmentation methods. Experiments further reveal a strong correlation between image generation quality and augmentation effectiveness. This framework offers a robust solution for data scarcity in cardiac image analysis, directly benefiting clinical applications. Full article

Objectives: Pelvic tumor resections demand high surgical precision to ensure clear margins while preserving function. Although patient-specific instruments (PSIs) improve osteotomy accuracy, positioning errors remain a limitation. This study evaluates the feasibility, accuracy, and usability of a novel dual-function augmented reality (AR) system for intraoperative guidance in PSI positioning and osteotomy execution using a head-mounted display (HMD). The system provides dual-function support by assisting both PSI placement and osteotomy execution. Methods: Ten fresh-frozen cadaveric hemipelves underwent AR-assisted internal hemipelvectomy, using customized 3D-printed PSIs and a new in-house AR software integrated into an HMD. Angular and translational deviations between planned and executed osteotomies were measured using postoperative CT analysis. Absolute angular errors were computed from plane normals; translational deviation was assessed as maximum error at the osteotomy corner point in both sagittal (pitch) and coronal (roll) planes. A Wilcoxon signed-rank test and Bland–Altman plots were used to assess intra-workflow cumulative error. Results: The mean absolute angular deviation was 5.11 ± 1.43°, with 86.66% of osteotomies within acceptable thresholds. Maximum pitch and roll deviations were 4.53 ± 1.32 mm and 2.79 ± 0.72 mm, respectively, with 93.33% and 100% of osteotomies meeting translational accuracy criteria. Wilcoxon analysis showed significantly lower angular error when comparing final executed planes to intermediate AR-displayed planes (p < 0.05), supporting improved PSI positioning accuracy with AR guidance. Surgeons rated the system highly (mean satisfaction ≥ 4.0) for usability and clinical utility. Conclusions: This cadaveric study confirms the feasibility and precision of an HMD-based AR system for PSI-guided pelvic osteotomies. The system demonstrated strong accuracy and high surgeon acceptance, highlighting its potential for clinical adoption in complex oncologic procedures. Full article

Accurate segmentation in medical imaging is essential for disease diagnosis and monitoring, particularly in lung imaging using proton and hyperpolarized gas MRI. However, image degradation due to noise and artifacts—especially in hyperpolarized gas MRI, where scans are acquired during breath-holds—poses challenges for conventional segmentation algorithms. This study evaluates the robustness of deep learning segmentation models under varying Gaussian noise levels, comparing traditional convolutional neural networks (CNNs) with modern Vision Transformer (ViT)-based models. Using a dataset of proton and hyperpolarized gas MRI slices from 56 participants, we trained and tested Feature Pyramid Network (FPN) and U-Net architectures with both CNN (VGG16, VGG19, ResNet152) and ViT (MiT-B0, B3, B5) backbones. Results showed that ViT-based models, particularly those using the SegFormer backbone, consistently outperformed CNN-based counterparts across all metrics and noise levels. The performance gap was especially pronounced in high-noise conditions, where transformer models retained higher Dice scores and lower boundary errors. These findings highlight the potential of ViT-based architectures for deployment in clinically realistic, low-SNR environments such as hyperpolarized gas MRI, where segmentation reliability is critical. Full article

Four-dimensional (4D) flow MRI has shown promise for the assessment of aortic hemodynamics. However, data analysis traditionally requires manual and time-consuming human input at several stages. This limits reproducibility and affects analysis workflows, such that large-cohort 4D flow studies are lacking. Here, a fully automated artificial intelligence (AI) 4D flow analysis pipeline was developed and evaluated in a cohort of over 350 subjects. The 4D flow MRI analysis pipeline integrated a series of previously developed and validated deep learning networks, which replaced traditionally manual processing tasks (background-phase correction, noise masking, velocity anti-aliasing, aorta 3D segmentation). Hemodynamic parameters (global aortic pulse wave velocity (PWV), peak velocity, flow energetics) were automatically quantified. The pipeline was evaluated in a heterogeneous single-center cohort of 379 subjects (age = 43.5 ± 18.6 years, 118 female) who underwent 4D flow MRI of the thoracic aorta (n = 147 healthy controls, n = 147 patients with a bicuspid aortic valve [BAV], n = 10 with mechanical valve prostheses, n = 75 pediatric patients with hereditary aortic disease). Pipeline performance with BAV and control data was evaluated by comparing to manual analysis performed by two human observers. A fully automated 4D flow pipeline analysis was successfully performed in 365 of 379 patients (96%). Pipeline-based quantification of aortic hemodynamics was closely correlated with manual analysis results (peak velocity: r = 1.00, p < 0.001; PWV: r = 0.99, p < 0.001; flow energetics: r = 0.99, p < 0.001; overall r ≥ 0.99, p < 0.001). Bland–Altman analysis showed close agreement for all hemodynamic parameters (bias 1–3%, limits of agreement 6–22%). Notably, limits of agreement between different human observers’ quantifications were moderate (4–20%). In addition, the pipeline 4D flow analysis closely reproduced hemodynamic differences between age-matched adult BAV patients and controls (median peak velocity: 1.74 m/s [automated] or 1.76 m/s [manual] BAV vs. 1.31 [auto.] vs. 1.29 [manu.] controls, p < 0.005; PWV: 6.4–6.6 m/s all groups, any processing [no significant differences]; kinetic energy: 4.9 μJ [auto.] or 5.0 μJ [manu.] BAV vs. 3.1 μJ [both] control, p < 0.005). This study presents a framework for the complete automation of quantitative 4D flow MRI data processing with a failure rate of less than 5%, offering improved measurement reliability in quantitative 4D flow MRI. Future studies are warranted to reduced failure rates and evaluate pipeline performance across multiple centers. Full article