Eur J Trauma Emerg Surg. 2025 Feb 20;51(1):115. doi: 10.1007/s00068-025-02779-w.

ABSTRACT

OBJECTIVES: Pediatric elbow fractures are a common injury among children. Recent advancements in artificial intelligence (AI), particularly deep learning (DL), have shown promise in diagnosing these fractures. This study systematically evaluated the performance of DL models in detecting pediatric elbow fractures.

MATERIALS AND METHODS: A comprehensive search was conducted in PubMed (Medline), EMBASE, and IEEE Xplore for studies published up to October 20, 2023. Studies employing DL models for detecting elbow fractures in patients aged 0 to 16 years were included. Key performance metrics, including sensitivity, specificity, and area under the curve (AUC), were extracted. The study was registered in PROSPERO (ID: CRD42023470558).

RESULTS: The search identified 22 studies, of which six met the inclusion criteria for the meta-analysis. The pooled sensitivity of DL models for pediatric elbow fracture detection was 0.93 (95% CI: 0.91-0.96). Specificity values ranged from 0.84 to 0.92 across studies, with a pooled estimate of 0.89 (95% CI: 0.85-0.92). The AUC ranged from 0.91 to 0.99, with a pooled estimate of 0.95 (95% CI: 0.93-0.97). Further analysis highlighted the impact of preprocessing techniques and the choice of model backbone architecture on performance.

CONCLUSION: DL models demonstrate exceptional accuracy in detecting pediatric elbow fractures. For optimal performance, we recommend leveraging backbone architectures like ResNet, combined with manual preprocessing supervised by radiology and orthopedic experts.

PMID:39976732 | DOI:10.1007/s00068-025-02779-w

Eur J Nucl Med Mol Imaging. 2025 Feb 20. doi: 10.1007/s00259-025-07145-x. Online ahead of print.

ABSTRACT

BACKGROUND: Myocardial perfusion imaging (MPI) using single-photon emission computed tomography (SPECT) is a well-established modality for noninvasive diagnostic assessment of coronary artery disease (CAD). However, the time-consuming and experience-dependent visual interpretation of SPECT images remains a limitation in the clinic.

PURPOSE: We aimed to develop advanced models to diagnose CAD using different supervised and semi-supervised deep learning (DL) algorithms and training strategies, including transfer learning and data augmentation, with SPECT-MPI and invasive coronary angiography (ICA) as standard of reference.

MATERIALS AND METHODS: A total of 940 patients who underwent SPECT-MPI were enrolled (281 patients included ICA). Quantitative perfusion SPECT (QPS) was used to extract polar maps of rest and stress states. We defined two different tasks, including (1) Automated CAD diagnosis with expert reader (ER) assessment of SPECT-MPI as reference, and (2) CAD diagnosis from SPECT-MPI based on reference ICA reports. In task 2, we used 6 strategies for training DL models. We implemented 13 different DL models along with 4 input types with and without data augmentation (WAug and WoAug) to train, validate, and test the DL models (728 models). One hundred patients with ICA as standard of reference (the same patients in task 1) were used to evaluate models per vessel and per patient. Metrics, such as the area under the receiver operating characteristics curve (AUC), accuracy, sensitivity, specificity, precision, and balanced accuracy were reported. DeLong and pairwise Wilcoxon rank sum tests were respectively used to compare models and strategies after 1000 bootstraps on the test data for all models. We also compared the performance of our best DL model to ER's diagnosis.

RESULTS: In task 1, DenseNet201 Late Fusion (AUC = 0.89) and ResNet152V2 Late Fusion (AUC = 0.83) models outperformed other models in per-vessel and per-patient analyses, respectively. In task 2, the best models for CAD prediction based on ICA were Strategy 3 (a combination of ER- and ICA-based diagnosis in train data), WoAug InceptionResNetV2 EarlyFusion (AUC = 0.71), and Strategy 5 (semi-supervised approach) WoAug ResNet152V2 EarlyFusion (AUC = 0.77) in per-vessel and per-patient analyses, respectively. Moreover, saliency maps showed that models could be helpful for focusing on relevant spots for decision making.

CONCLUSION: Our study confirmed the potential of DL-based analysis of SPECT-MPI polar maps in CAD diagnosis. In the automation of ER-based diagnosis, models' performance was promising showing accuracy close to expert-level analysis. It demonstrated that using different strategies of data combination, such as including those with and without ICA, along with different training methods, like semi-supervised learning, can increase the performance of DL models. The proposed DL models could be coupled with computer-aided diagnosis systems and be used as an assistant to nuclear medicine physicians to improve their diagnosis and reporting, but only in the LAD territory.

CLINICAL TRIAL NUMBER: Not applicable.

PMID:39976703 | DOI:10.1007/s00259-025-07145-x

J Appl Clin Med Phys. 2025 Feb 20:e70031. doi: 10.1002/acm2.70031. Online ahead of print.

ABSTRACT

PURPOSE: To compare image quality and clinical utility of a T2-weighted (T2W) 3-dimensional (3D) fast spin echo (FSE) sequence using deep learning reconstruction (DLR) versus conventional reconstruction for rectal magnetic resonance imaging (MRI).

METHODS: The study included 50 patients with rectal cancer who underwent rectal MRI consecutively between July 7, 2020 and January 20, 2021 using a T2W 3D FSE sequence with DLR and conventional reconstruction. Three radiologists reviewed the two sets of images, scoring overall SNR, motion artifacts, and overall image quality on a 3-point scale and indicating clinical preference for DLR or conventional reconstruction based on those three criteria as well as image characterization of bowel wall layer definition, tumor invasion of muscularis propria, residual disease, fibrosis, nodal margin, and extramural venous invasion.

RESULTS: Image quality was rated as moderate or good for both DLR and conventional reconstruction for most cases. DLR was preferred over conventional reconstruction in all of the categories except for bowel wall layer definition.

CONCLUSION: Both conventional reconstruction and DLR provide acceptable image quality for T2W 3D FSE imaging of rectal cancer. DLR was clinically preferred over conventional reconstruction in almost all categories.

PMID:39976552 | DOI:10.1002/acm2.70031

Med Phys. 2025 Feb 20. doi: 10.1002/mp.17696. Online ahead of print.

ABSTRACT

BACKGROUND: Deformable medical image registration aims to align image pairs with local differences, improving the accuracy of medical analyses and assisting various diagnostic scenarios.

PURPOSE: We aim to overcome these challenges: Deep learning-based registration approaches have greatly enhanced registration speed and accuracy by continuously improving registration networks and processes. However, the lack of extensive medical datasets limits the complexity of registration models. Optimizing registration networks within a fixed dataset often leads to overfitting, hindering further accuracy improvements and reducing generalization capabilities.

METHODS: We explore the application of the foundational model DINOv2 to registration tasks, leveraging its prior knowledge to support learning-based unsupervised registration networks and overcome network bottlenecks to improve accuracy. We investigate three modes of DINOv2-assisted registration, including direct registration architecture, enhanced architecture, and refined architecture. Additionally, we study the applicability of three feature aggregation methods-convolutional interaction, direct fusion, and cross-attention-within the proposed DINOv2-based registration frameworks.

RESULTS: We conducted extensive experiments on the IXI and OASIS public datasets, demonstrating that the enhanced and refined architectures notably improve registration accuracy, reduce data dependency, and maintain strong generalization capabilities.

CONCLUSION: This study offers novel approaches for applying foundational models to deformable image registration tasks.

PMID:39976314 | DOI:10.1002/mp.17696

Stat Med. 2025 Feb 28;44(5):e70020. doi: 10.1002/sim.70020.

ABSTRACT

Freezing of Gait (FOG) is a prevalent symptom in advanced Parkinson's Disease (PD), characterized by intermittent transitions between normal gait and freezing episodes. This study introduces a novel graph-theoretic approach to detect FOG from video data of PD patients. We construct a sequence of pose graphs that represent the spatial relations and temporal progression of a patient's posture over time. Each graph node corresponds to an estimated joint position, while the edges reflect the anatomical connections and their proximity. We propose a hypothesis testing procedure that deploys the Fréchet statistics to identify break points in time between regular gait and FOG episodes, where we model the central tendency and dispersion of the pose graphs in the presentation of graph Laplacian matrices by computing their Fréchet mean and variance. We implement binary segmentation and incremental computation in our algorithm for efficient calculation. The proposed framework is validated on two datasets, Kinect3D and AlphaPose, demonstrating its effectiveness in detecting FOG from video data. The proposed approach that extracts matrix features is distinct from the prevailing pixel-based deep learning methods. It provides a new perspective on feature extraction for FOG detection and potentially contributes to improved diagnosis and treatment of PD.

PMID:39976295 | DOI:10.1002/sim.70020

ArXiv [Preprint]. 2025 Feb 1:arXiv:2501.14158v2.

ABSTRACT

Magnetic resonance imaging (MRI) is a non-invasive imaging modality and provides comprehensive anatomical and functional insights into the human body. However, its long acquisition times can lead to patient discomfort, motion artifacts, and limiting real-time applications. To address these challenges, strategies such as parallel imaging have been applied, which utilize multiple receiver coils to speed up the data acquisition process. Additionally, compressed sensing (CS) is a method that facilitates image reconstruction from sparse data, significantly reducing image acquisition time by minimizing the amount of data collection needed. Recently, deep learning (DL) has emerged as a powerful tool for improving MRI reconstruction. It has been integrated with parallel imaging and CS principles to achieve faster and more accurate MRI reconstructions. This review comprehensively examines DL-based techniques for MRI reconstruction. We categorize and discuss various DL-based methods, including end-to-end approaches, unrolled optimization, and federated learning, highlighting their potential benefits. Our systematic review highlights significant contributions and underscores the potential of DL in MRI reconstruction. Additionally, we summarize key results and trends in DL-based MRI reconstruction, including quantitative metrics, the dataset, acceleration factors, and the progress of and research interest in DL techniques over time. Finally, we discuss potential future directions and the importance of DL-based MRI reconstruction in advancing medical imaging. To facilitate further research in this area, we provide a GitHub repository that includes up-to-date DL-based MRI reconstruction publications and public datasets-https://github.com/mosaf/Awesome-DL-based-CS-MRI.

PMID:39975448 | PMC:PMC11838702

ArXiv [Preprint]. 2025 Jan 27:arXiv:2501.15712v1.

ABSTRACT

Computational modeling of cardiovascular function has become a critical part of diagnosing, treating and understanding cardiovascular disease. Most strategies involve constructing anatomically accurate computer models of cardiovascular structures, which is a multistep, time-consuming process. To improve the model generation process, we herein present SeqSeg (sequential segmentation): a novel deep learning based automatic tracing and segmentation algorithm for constructing image-based vascular models. SeqSeg leverages local U-Net-based inference to sequentially segment vascular structures from medical image volumes. We tested SeqSeg on CT and MR images of aortic and aortofemoral models and compared the predictions to those of benchmark 2D and 3D global nnU-Net models, which have previously shown excellent accuracy for medical image segmentation. We demonstrate that SeqSeg is able to segment more complete vasculature and is able to generalize to vascular structures not annotated in the training data.

PMID:39975447 | PMC:PMC11838707

ArXiv [Preprint]. 2025 Jan 24:arXiv:2311.11046v2.

ABSTRACT

Major depressive disorder (MDD) is a complex psychiatric disorder that affects the lives of hundreds of millions of individuals around the globe. Even today, researchers debate if morphological alterations in the brain are linked to MDD, likely due to the heterogeneity of this disorder. The application of deep learning tools to neuroimaging data, capable of capturing complex non-linear patterns, has the potential to provide diagnostic and predictive biomarkers for MDD. However, previous attempts to demarcate MDD patients and healthy controls (HC) based on segmented cortical features via linear machine learning approaches have reported low accuracies. Here, we used globally representative data from the ENIGMA-MDD working group containing 7,012 participants from 30 sites (N=2,772 MDD and N=4,240 HC), which allows a comprehensive analysis with generalizable results. Based on the hypothesis that integration of vertex-wise cortical features can improve classification performance, we evaluated the classification of a DenseNet and a Support Vector Machine (SVM), with the expectation that the former would outperform the latter. We found that both classifiers exhibited close to chance performance (balanced accuracy DenseNet: 51%; SVM: 53%), when estimated on unseen sites. Slightly higher classification performance (balanced accuracy DenseNet: 58%; SVM: 55%) was found when the cross-validation folds contained subjects from all sites, indicating site effect. In conclusion, the integration of vertex-wise morphometric features and the use of the non-linear classifier did not lead to the differentiability between MDD and HC. Our results support the notion that MDD classification on this combination of such features and classifiers is unfeasible. Perhaps more sophisticated integration of multimodal information may lead to a higher performance in this diagnostic task.

PMID:39975425 | PMC:PMC11838705

bioRxiv [Preprint]. 2025 Feb 8:2025.02.07.637149. doi: 10.1101/2025.02.07.637149.

ABSTRACT

As spatial molecular data grow in scope and resolution, there is a pressing need to identify key spatial structures associated with disease. Current approaches often rely on hand-crafted features such as local abundances of manually annotated, discrete cell types, which may overlook important signals. Here we introduce variational inference-based microniche analysis (VIMA), a method that combines deep learning with principled statistics to discover associated spatial features with greater flexibility and precision. VIMA uses a variational autoencoder to extract numerical "fingerprints" from small tissue patches that capture their biological content. It uses these fingerprints to define a large number of "microniches" - small, potentially overlapping groups of tissue patches with highly similar biology that span multiple samples. It then uses rigorous statistics to identify microniches whose abundance correlates with case-control status. We show in simulations that VIMA is well calibrated and more powerful and accurate than other approaches. We then apply VIMA to a 140-gene spatial transcriptomics dataset in Alzheimer's dementia, a 54-marker CO-Detection by indEXing (CODEX) dataset in ulcerative colitis (UC), and a 7-marker immunohistochemistry dataset in rheumatoid arthritis (RA), in each case recapitulating known biology and identifying novel spatial features of disease.

PMID:39975274 | PMC:PMC11839118 | DOI:10.1101/2025.02.07.637149

bioRxiv [Preprint]. 2025 Jan 27:2025.01.24.634753. doi: 10.1101/2025.01.24.634753.

ABSTRACT

Given the successful graft-versus-leukemia cell treatment effect observed with allogeneic hematopoietic stem cell transplant for patients with refractory or relapsed acute myeloid leukemia, immunotherapies have also been investigated in the nontransplant setting. Here, we use a multi-omic approach to investigate spatiotemporal interactions in the bone marrow niche between leukemia cells and immune cells in patients with refractory or relapsed acute myeloid leukemia treated with a combination of the immune checkpoint inhibitor pembrolizumab and hypomethylating agent decitabine. We derived precise segmentation data by extensively training nuclear and membrane cell segmentation models, which enabled accurate transcript assignment and deep learning-feature-based image analysis. To overcome read-depth limitations, we integrated the single-cell RNA sequencing data with single-cell-resolution spatial transcriptomic data from the same sample. Quantifying cell-cell distances between cell edges rather than cell centroids allowed us to conduct a more accurate downstream analysis of the tumor microenvironment, revealing that multiple cell types of interest had global enrichment or local enrichment proximal to leukemia cells after pembrolizumab treatment, which could be associated with their clinical responses. Furthermore, ligand-receptor analysis indicated a potential increase in TWEAK signaling between leukemia cells and immune cells after pembrolizumab treatment.

HIGHLIGHTS: Spatial transcriptomic analysis of R-AML bone marrow niches provides detailed information about intercellular interactions in the tumor microenvironment.Immunotherapy shifts the cell composition of the leukemia neighborhood.

PMID:39975227 | PMC:PMC11838223 | DOI:10.1101/2025.01.24.634753

bioRxiv [Preprint]. 2025 Jan 29:2025.01.29.634860. doi: 10.1101/2025.01.29.634860.

ABSTRACT

Identification of neuron type is critical to understand computation in neural circuits through extracellular recordings in awake, behaving animal subjects. Yet, modern recording probes have limited power to resolve neuron type. Here, we leverage the well-characterized architecture of the cerebellar circuit to perform expert identification of neuron type from extracellular recordings in behaving non-human primates. Using deep-learning classifiers we evaluate the information contained in readily accessible extracellular features for neuron identification. Waveform, discharge statistics, anatomical layer, and functional interactions each can inform neuron labels for a sizable fraction of cerebellar units. Together, as inputs to a deep-learning classifier, the features perform even better. Our tools and methodologies, validated during smooth pursuit eye movements in the cerebellar floccular complex of awake behaving monkeys, can guide expert identification of neuron type during cerebellar-dependent tasks in behaving animals across species. They lay the groundwork for characterization of information processing in the cerebellar cortex.

IMPACT STATEMENT: To understand how the brain performs computations in the service of behavior, we develop methods to link neuron type to functional activity within well-characterized neural circuits. Here, we show how features derived from extracellular recordings provide complementary information to disambiguate neuron identity in the cerebellar cortex.

PMID:39975199 | PMC:PMC11838295 | DOI:10.1101/2025.01.29.634860

bioRxiv [Preprint]. 2025 Feb 5:2025.01.31.635598. doi: 10.1101/2025.01.31.635598.

ABSTRACT

Deep learning excels at segmenting objects within noisy cryo-electron tomograms, but the approach is typically bottlenecked by access to ground truth training data. To address this issue we have developed CryoTomoSim (CTS), an open-source software package that builds coarse-grained models of macromolecular complexes embedded in vitreous ice and then simulates transmitted electron tilt series for tomographic reconstruction. Using CTS outputs, we demonstrate the effects of key microscope parameters (dose, defocus, and pixel size) on deep learning-based segmentation, and show that including both molecular crowding and diversity within synthetic datasets is key to training cellular segmentation networks from purely synthetic inputs. While very effective as initial models, the accuracy of these networks is currently limited, and real cellular data is necessary to train the most accurate and generalizable U-Nets. Using a co-training approach, we first segment over 100 tomograms from neuronal growth cones to quantify their cytoskeletal distributions and then we build a generalized cellular cryo-ET segmentation network called NeuralSeg that can segment a subset of cellular features in tomograms from all domains of life.

PMID:39975172 | PMC:PMC11838407 | DOI:10.1101/2025.01.31.635598

bioRxiv [Preprint]. 2025 Jan 30:2025.01.29.635570. doi: 10.1101/2025.01.29.635570.

ABSTRACT

Spatial correlation of functional connectivity profiles across matching anatomical locations in individuals is often calculated to delineate individual differences in functional networks. Likewise, spatial correlation is assessed across average functional connectivity profiles of groups to evaluate the maturity of functional networks during development. Despite its widespread use, spatial correlation is limited to comparing two samples at a time. In this study, we employed a variational autoencoder to embed functional connectivity profiles from various anatomical locations, individuals, and group averages for simultaneous comparison. We demonstrate that our variational autoencoder, with pre-trained weights, can project new functional connectivity profiles from the vertex space to a latent space with as few as two dimensions, yet still retain meaningful global and local structures in the data. Functional connectivity profiles from various functional networks occupy distinct compartments of the latent space. Moreover, the variability of functional connectivity profiles from the same anatomical location is readily captured in the latent space. We believe that this approach could be useful for visualization and exploratory analyses in precision functional mapping.

PMID:39975052 | PMC:PMC11838398 | DOI:10.1101/2025.01.29.635570

bioRxiv [Preprint]. 2025 Feb 5:2024.08.29.610300. doi: 10.1101/2024.08.29.610300.

ABSTRACT

The proto-oncogene Ras which governs diverse intracellular pathways has four major isoforms (KRAS4A, KRAS4B, HRAS, and NRAS) with substantial sequence homology and similar in vitro biochemistry. There is considerable interest in investigating the roles of these independently as their association with different cancers vary, but there are few Ras isoform-specific binding reagents as the only significant sequence differences are in their disordered and highly charged C-termini which have been difficult to elicit antibodies against. To overcome this limitation, we use deep learning-based methods to de novo design Ras isoform-specific binders (RIBs) for all major Ras isoforms that specifically target the Ras C-terminus. The RIBs bind to their target Ras isoforms both in vitro and in cells with remarkable specificity, disrupting their membrane localization and inhibiting Ras activity, and should contribute to dissecting the distinct roles of Ras isoforms in biology and disease.

PMID:39975043 | PMC:PMC11838417 | DOI:10.1101/2024.08.29.610300

bioRxiv [Preprint]. 2025 Feb 5:2025.02.04.636539. doi: 10.1101/2025.02.04.636539.

ABSTRACT

The development of therapies and vaccines targeting integral membrane proteins has been complicated by their extensive hydrophobic surfaces, which can make production and structural characterization difficult. Here we describe a general deep learning-based design approach for solubilizing native membrane proteins while preserving their sequence, fold, and function using genetically encoded de novo protein WRAPs ( W ater-soluble R Fdiffused A mphipathic P roteins) that surround the lipid-interacting hydrophobic surfaces, rendering them stable and water-soluble without the need for detergents. We design WRAPs for both beta-barrel outer membrane and helical multi-pass transmembrane proteins, and show that the solubilized proteins retain the binding and enzymatic functions of the native targets with enhanced stability. Syphilis vaccine development has been hindered by difficulties in characterizing and producing the outer membrane protein antigens; we generated soluble versions of four Treponema pallidum outer membrane beta barrels which are potential syphilis vaccine antigens. A 4.0 Å cryo-EM map of WRAPed TP0698 is closely consistent with the design model. WRAPs should be broadly useful for facilitating biochemical and structural characterization of integral membrane proteins, enabling therapeutic discovery by screening against purified soluble targets, and generating antigenically intact immunogens for vaccine development.

PMID:39975033 | PMC:PMC11838538 | DOI:10.1101/2025.02.04.636539

Sci Rep. 2025 Feb 19;15(1):6011. doi: 10.1038/s41598-025-89651-4.

ABSTRACT

Traffic congestion, particularly in rapidly expanding urban centers, significantly impacts the timely delivery of emergency medical services (EMS), where every minute can mean the difference between life and death. Traditional traffic signal control systems often lack real-time adaptability to prioritize emergency vehicles, resulting in delays caused by congestion around ambulances. To address this critical issue, this paper presents an AI-driven real-time traffic management system designed to reduce EMS response times. The proposed solution incorporates three core components: Raspberry Pi-based traffic signal prioritization, deep learning-enabled audio-visual ambulance detection, and an advanced intelligent traffic management framework. For audio detection, raw data is transformed into spectrograms using Mel Frequency Cepstral Coefficients (MFCCs) and classified using a Long Short-Term Memory (LSTM) network. Visual data is processed through a ResNet18 convolutional neural network, pre-trained on ImageNet using inductive transfer learning. The outputs from the auditory and visual streams are integrated using empirical risk minimization, enabling accurate ambulance detection through multimodal data fusion. Performance evaluation demonstrates the effectiveness of the proposed system, achieving 98.3% accuracy in audio classification, 98.1% accuracy in visual classification, and 99% accuracy with the fused model. Additional metrics, including precision, recall, F1-score, and a confusion matrix, confirm the model's reliability. This innovative system has the potential to transform urban traffic networks into intelligent, adaptive systems, reducing delays caused by traffic congestion, enhancing emergency medical care response times, and ultimately saving lives. The framework offers a scalable blueprint for future smart city traffic management solutions, meticulously designed to support urban growth and expansion.

PMID:39971977 | DOI:10.1038/s41598-025-89651-4

Sci Rep. 2025 Feb 19;15(1):6035. doi: 10.1038/s41598-025-89348-8.

ABSTRACT

Diet, stress, genetics, and a sedentary lifestyle may all contribute to heart disease rates. Although recent studies propose comprehensive automated diagnostic systems, these systems tend to focus on one aspect, such as feature selection, prioritization, or predictive accuracy. A more complete approach that considers all of these factors can improve the efficiency of a cardiac prediction system. This study uses an appropriate strategy to overcome potential network design problems, design challenges, overfitting, and lack of robustness that can interfere with system performance. The research introduces an ideally designed deep trust network called ID-DTN to improve system performance. The Ruzzo-Tompa method is used to eliminate noncontributory features. The Seagull Optimization Algorithm (SOA) is introduced to optimize the trust depth network to achieve optimal network design. The study scrutinizes the deep trust network (ID-DTN) and the restricted Boltzmann machine (RBM) and sheds light on the system's operation. This proposal can optimize both network architecture and feature selection, which is the main novelty. The proposed method is analyzed using the below-mentioned metrics: Matthew's correlation coefficient, F1 score, accuracy, sensitivity, specificity, and accuracy. ID-DTN performs well compared to other state-of-the-art methods. The validation results confirm that the proposed method improves the prediction accuracy to 97.11% and provides reliable recommendations for patients with cardiovascular disease.

PMID:39971944 | DOI:10.1038/s41598-025-89348-8

J Imaging Inform Med. 2025 Feb 19. doi: 10.1007/s10278-025-01440-7. Online ahead of print.

ABSTRACT

Segmentation and classification of breast tumors are two critical tasks since they provide significant information for computer-aided breast cancer diagnosis. Combining these tasks leverages their intrinsic relevance to enhance performance, but the variability and complexity of tumor characteristics remain challenging. We propose a novel multi-task deep learning network (NMTNet) for the joint segmentation and classification of breast tumors, which is based on a convolutional neural network (CNN) and U-shaped architecture. It mainly comprises a shared encoder, a multi-scale fusion channel refinement (MFCR) module, a segmentation branch, and a classification branch. First, ResNet18 is used as the backbone network in the encoding part to enhance the feature representation capability. Then, the MFCR module is introduced to enrich the feature depth and diversity. Besides, the segmentation branch combines a lesion region enhancement (LRE) module between the encoder and decoder parts, aiming to capture more detailed texture and edge information of irregular tumors to improve segmentation accuracy. The classification branch incorporates a fine-grained classifier that reuses valuable segmentation information to discriminate between benign and malignant tumors. The proposed NMTNet is evaluated on both ultrasound and magnetic resonance imaging datasets. It achieves segmentation dice scores of 90.30% and 91.50%, and Jaccard indices of 84.70% and 88.10% for each dataset, respectively. And the classification accuracy scores are 87.50% and 99.64% for the corresponding datasets, respectively. Experimental results demonstrate the superiority of NMTNet over state-of-the-art methods on breast tumor segmentation and classification tasks.

PMID:39971818 | DOI:10.1007/s10278-025-01440-7

J Imaging Inform Med. 2025 Feb 19. doi: 10.1007/s10278-025-01446-1. Online ahead of print.

ABSTRACT

The rapid advancements of deep learning technology have revolutionized medical imaging diagnosis. However, training these models is often challenged by label imbalance and the scarcity of certain diseases. Most models fail to recognize multiple coexisting diseases, which are common in real-world clinical scenarios. Moreover, most radiological models rely solely on image data, which contrasts with radiologists' comprehensive approach, incorporating both images and other clinical information such as clinical history and laboratory results. In this study, we introduce a Multimodal Chest X-ray Network (MCX-Net) that integrates chest X-ray images and clinical history texts for multi-label disease diagnosis. This integration is achieved by combining a pretrained text encoder, a pretrained image encoder, and a pretrained image-text cross-modal encoder, fine-tuned on the public MIMIC-CXR-JPG dataset, to diagnose 13 diverse lung diseases on chest X-rays. As a result, MCX-Net achieved the highest macro AUROC of 0.816 on the test set, significantly outperforming unimodal baselines such as ViT-base and ResNet152, which scored 0.747 and 0.749, respectively (p < 0.001). This multimodal approach represents a substantial advancement over existing image-based deep-learning diagnostic systems for chest X-rays.

PMID:39971817 | DOI:10.1007/s10278-025-01446-1

BJOG. 2025 Feb 19. doi: 10.1111/1471-0528.18097. Online ahead of print.

ABSTRACT

Cardiotocography (CTG), introduced in the 1960s, was initially expected to prevent hypoxia-related deaths and neurological injuries. However, more than five decades later, evidence supporting the evidence of intrapartum CTG in preventing neonatal and long-term childhood morbidity and mortality remains inconclusive. At the same time, shortcomings in CTG interpretation have been recognised as important contributory factors to rising caesarean section rates and missed opportunities for timely interventions. An important limitation is its high false-positive rate and poor specificity, which undermines reliably identifying foetuses at risk of hypoxia-related injuries. These shortcomings are compounded by the technology's significant intra- and interobserver variability, as well as the subjective and complex nature of fetal heart rate interpretation. However, human factors and other environmental factors are equally significant. Advancements in fetal heart rate monitoring are crucial to support clinicians in improving health outcomes for newborns and their mothers, while at the same time avoiding unnecessary operative deliveries. These limitations highlight the clinical need to enhance neonatal outcomes while minimising unnecessary interventions, such as instrumental deliveries or caesarean sections. We believe that achieving this requires a paradigm shift from subjective interpretation of complex and nonspecific fetal heart rate patterns to evidence-based, quantifiable solutions that integrate hardware, engineering and clinical perspectives. Such transformation necessitates an international, multidisciplinary effort encompassing the entire continuum of pregnancy care and the broader healthcare ecosystem, with emphasis on well-defined, actionable health outcomes. Achieving this will depend on collaborations between researchers, clinicians, medical device manufacturers and other relevant stakeholders. This expert review paper outlines the most relevant and promising directions for research and strategic initiatives to address current challenges in fetal heart rate monitoring. Key themes include advancements in computerised fetal heart rate monitoring, the application of big data and artificial intelligence, innovations in home and remote monitoring and consideration of human factors.

PMID:39971749 | DOI:10.1111/1471-0528.18097