BMC Biol. 2025 Jul 15;23(1):212. doi: 10.1186/s12915-025-02314-8.

ABSTRACT

BACKGROUND: Accurate identification of anti-inflammatory peptides (AIPs) is crucial for drug development and inflammatory disease treatment. However, the short length and limited informational content of peptide sequences make precise computational recognition particularly challenging. While various machine learning and deep learning approaches have been explored, their limitations in feature representation and model integration hinder the effective discovery of novel AIPs.

RESULTS: In this study, we present NeXtMD, a novel dual-module stacked framework that integrates both machine learning (ML) and deep learning (DL) components for accurate AIP identification. NeXtMD systematically extracts four functionally relevant sequence-derived descriptors-residue composition, inter-residue correlation, physicochemical properties, and sequence patterns-and utilizes a two-stage prediction strategy. The first stage generates preliminary predictions using four distinct encoding strategies and ML classifiers, while the second stage employs a multi-branch residual network (ResNeXt) to refine prediction outputs. Benchmark evaluations demonstrate that NeXtMD outperforms current state-of-the-art methods on multiple performance metrics. Moreover, NeXtMD maintains strong generalization capabilities when applied to unseen peptide sequences, showing its robustness and scalability.

CONCLUSIONS: NeXtMD offers a high-performance and interpretable computational framework for AIP identification, with significant potential to facilitate the discovery and design of peptide-based anti-inflammatory therapeutics. The architecture and methodological innovations of NeXtMD also provide a generalizable strategy that can be adapted to other bioactive peptide prediction tasks, supporting broader applications in therapeutic peptide development.

PMID:40660190 | DOI:10.1186/s12915-025-02314-8

Ann Nucl Med. 2025 Jul 14. doi: 10.1007/s12149-025-02081-0. Online ahead of print.

ABSTRACT

PURPOSE: Accurate differentiation of Parkinsonism subtypes-including Parkinson's disease (PD), multiple system atrophy (MSA), and progressive supranuclear palsy (PSP)-is essential for clinical prognosis and treatment planning. However, this remains a major challenge due to overlapping symptomatology and high inter-individual variability in cerebral glucose metabolism patterns observed on fluorodeoxyglucose positron emission tomography (FDG-PET).

METHODS: To address these challenges, we propose PETFormer-SCL, a clinically informed deep learning framework that integrates convolutional neural networks (CNNs) with a channel-wise Transformer module, guided by supervised contrastive learning (SCL). This architecture is designed to enhance disease-specific feature learning while mitigating individual variability.

RESULTS: Trained on 945 patients and evaluated on an independent test cohort of 330 patients (1275 in total), PETFormer-SCL achieved AUCs of 0.9830, 0.9702, and 0.9565 for MSA, PD, and PSP, respectively. In addition, class activation maps (CAMs) highlighted key disease-related brain regions-including the cerebellum, midbrain, and basal ganglia-demonstrating strong alignment with known pathophysiological findings.

CONCLUSIONS: PETFormer-SCL not only achieves high diagnostic accuracy, particularly for subtypes with overlapping phenotypes, but also enhances interpretability. These results support its potential as a reliable clinical decision-support tool for the early and differential diagnosis of Parkinsonism.

PMID:40660058 | DOI:10.1007/s12149-025-02081-0

Aesthetic Plast Surg. 2025 Jul 14. doi: 10.1007/s00266-025-05068-4. Online ahead of print.

ABSTRACT

BACKGROUND: Driven by advancements in deep learning, surgical robots, and predictive modeling technologies, the integration of artificial intelligence (AI) and plastic surgery has expanded rapidly. Although AI shows the potential to enhance precision and efficiency, its clinical integration faces challenges, including ethical concerns and interdisciplinary complexity, which require a systematic analysis of research trends.

METHODS: The CiteSpace and VOSviewer software were used to conduct a quantitative analysis of 235 documents in the core collection of Web of Science from 2016 to 2024. Co-citation networks, keyword co-occurrence, burst detection, and cluster analysis were employed to map the research trajectories. The inclusion criteria gave priority to studies that explicitly incorporated artificial intelligence into surgical designs or outcomes. The contributions of countries, institutions, and authors were evaluated through centrality indicators.

RESULT: Publications related to artificial intelligence have grown exponentially, with the USA, Germany, and Canada leading research output. Harvard and Stanford Universities dominate in terms of institutional contributions, but cross-institutional collaboration remains limited. The keyword cluster highlights the innovations of artificial intelligence in breast reconstruction, facial analysis, and automated grading systems. Burst terms such as "deep learning," "risk assessment," and "attractiveness" underscore AI's role in optimizing surgical outcomes, but they also expose biases against Western-centric beauty standards. Ethical concerns, dataset diversity gaps, and overreliance on AI-driven decisions have become key obstacles.

CONCLUSION: The integration of artificial intelligence in plastic surgery goes beyond the utility based on tools and into data-informed surgical engineering. The persistent gap in collaboration and dataset diversity highlights the need for global, interdisciplinary efforts to address technical and ethical challenges while advancing AI's clinical utility. Future research must prioritize transparency, inclusivity, and collaborative innovation to realize AI's transformative potential while mitigating risks.

LEVEL OF EVIDENCE IV: This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .

PMID:40660032 | DOI:10.1007/s00266-025-05068-4

J Imaging Inform Med. 2025 Jul 14. doi: 10.1007/s10278-025-01594-4. Online ahead of print.

ABSTRACT

As monkeypox is spreading rapidly, the incidence of monkeypox has been increasing in recent times. Therefore, it is very important to detect and diagnose this disease to get effective treatment planning. The prominent aim of this paper is to design an effective monkeypox detection and classification model by utilizing deep learning and classification models. In this study, a novel Adaptive Generative 3D VNet model is presented to effectively classify the monkeypox lesions. Different data augmentation approaches, deep learning, and adaptive fusion are integrated into the proposed model to attain better results in disease classification. The major objective of the proposed model is to mitigate the challenges of limited labeled data by generating synthetic augmented images and combining them with real images for robust classification. The two major components of the proposed system are the Adaptive Generative Network and the 3D VNet. Additional training models are generated by the adaptive generative network through augmentation approaches including cropping, rotation, and flipping, thereby increasing the diversity of the dataset. The 3D VNet processes these images in a volumetric manner to capture spatial relationships within the lesions, improving classification accuracy. The fusion layer then adaptively combines the predictions from the real and augmented data to optimize the overall effectiveness of a model. Key performance metrics including accuracy, precision, sensitivity, specificity, Jaccard Index, Hausdorff distance, and Dice Similarity Coefficient are used to compute the effectiveness of a model. The findings show that the Adaptive Generative 3D VNet model outperforms traditional 2D models by significantly improving the classification accuracy and robustness, especially in the presence of limited labeled data. Therefore, the simulation results demonstrate that the proposed model achieves high accuracy and precision of 98.8% and 98.5%, respectively based on the Monkeypox Skin Lesion Dataset.

PMID:40659969 | DOI:10.1007/s10278-025-01594-4

J Imaging Inform Med. 2025 Jul 14. doi: 10.1007/s10278-025-01608-1. Online ahead of print.

ABSTRACT

In this study, we proposed a lightweight neural network architecture based on inverted residual network, efficient squeeze excitation (ESE) module, and double transfer learning, called TLese-ResNet, for breast cancer molecular subtype recognition. The inverted ResNet reduces the number of network parameters while enhancing the cross-layer gradient propagation and feature expression capabilities. The introduction of the ESE module reduces the network complexity while maintaining the channel relationship collection. The dataset of this study comes from the mammography images of patients diagnosed with invasive breast cancer in a hospital in Jiangxi. The dataset comprises preoperative mammography images with CC and MLO views. Given that the dataset is somewhat small, in addition to the commonly used data augmentation methods, double transfer learning is also used. Double transfer learning includes the first transfer, in which the source domain is ImageNet and the target domain is the COVID-19 chest X-ray image dataset, and the second transfer, in which the source domain is the target domain of the first transfer, and the target domain is the mammography dataset we collected. By using five-fold cross-validation, the mean accuracy and area under received surgery feature on mammographic images of CC and MLO views were 0.818 and 0.883, respectively, outperforming other state-of-the-art deep learning-based models such as ResNet-50 and DenseNet-121. Therefore, the proposed model can provide clinicians with an effective and non-invasive auxiliary tool for molecular subtype identification of breast cancer.

PMID:40659967 | DOI:10.1007/s10278-025-01608-1

Urologie. 2025 Jul 14. doi: 10.1007/s00120-025-02651-0. Online ahead of print.

ABSTRACT

BACKGROUND: Urology presents itself as a modern and future-oriented discipline. Artificial intelligence (AI) is becoming increasingly important in the field of digital urology. It enables the analysis of large medical data sets, such as histological patterns, imaging or molecular markers. In addition, large language models (LLMs) are opening up new fields of application. Digital health applications (DiGA) are also becoming increasingly important.

OBJECTIVES: This article provides an overview of current developments in digital urology. The focus is on the possible applications of AI and LLMs, taking into account the legal and ethical framework, as well as an overview of the therapeutic landscape of DiGA in urological care.

METHODS: Existing AI models and applications will be presented and analyzed. In addition, DiGA in urological care and current regulatory requirements are discussed.

RESULTS: AI systems can improve many areas of urology-for example through automated evaluation of image morphology and histology data or voice-based applications in documentation and communication. DiGA support the follow-up care and treatment of urological diseases. Their use requires transparency, data protection, and quality assurance.

CONCLUSION: AI and digital applications offer new care models and can help to increase efficiency. Challenges exist in terms of data quality, information technology infrastructure and the safe and responsible use of technologies. Future developments should be interdisciplinary.

PMID:40659882 | DOI:10.1007/s00120-025-02651-0

JMIR Med Inform. 2025 Jul 14;13:e66973. doi: 10.2196/66973.

ABSTRACT

BACKGROUND: Standardized registries, such as the International Classification of Diseases (ICD) codes, are commonly built using administrative codes assigned to patient encounters. However, patients with fall injury are often coded using subsequent injury codes, such as hip fractures. This necessitates manual screening to ensure the accuracy of data registries.

OBJECTIVE: This study aimed to automate the extraction of fall incidents and mechanisms using natural language processing (NLP) and compare this approach with the ICD method.

METHODS: Clinical notes for patients with fall-induced hip fractures were retrospectively reviewed by medical experts. Fall incidences were detected, annotated, and classified among patients who had a fall-induced hip fracture (case group). The control group included patients with hip fractures without any evidence of falls. NLP models were developed using the annotated notes of the study groups to fulfill two separate tasks: fall occurrence detection and fall mechanism classification. The performances of the models were compared using accuracy, sensitivity, specificity, positive predictive value, negative predictive value, F1-score, and area under the receiver operating characteristic curve.

RESULTS: A total of 1769 clinical notes were included in the final analysis for the fall occurrence task, and 783 clinical notes were analyzed for the fall mechanism classification task. The highest F1-score using NLP for fall occurrence was 0.97 (specificity=0.96; sensitivity=0.97), and for fall mechanism classification was 0.61 (specificity=0.56; sensitivity=0.62). Natural language processing could detect up to 98% of the fall occurrences and 65% of the fall mechanisms accurately, compared to 26% and 12%, respectively, by ICD codes.

CONCLUSIONS: Our findings showed promising performance with higher accuracy of NLP algorithms compared to the conventional method for detecting fall occurrence and mechanism in developing disease registries using clinical notes. Our approach can be introduced to other registries that are based on large data and are in need of accurate annotation and classification.

PMID:40658984 | DOI:10.2196/66973

PLoS One. 2025 Jul 14;20(7):e0327733. doi: 10.1371/journal.pone.0327733. eCollection 2025.

ABSTRACT

BACKGROUND: Parkinson's disease (PD), a progressive neurodegenerative disorder prevalent in aging populations, manifests clinically through characteristic motor impairments including bradykinesia, rigidity, and resting tremor. Early detection and timely intervention may delay disease progression. Spiral drawing tasks have been established as effective auxiliary diagnostic tools. This study developed a hybrid deep learning model to analyze motion data from finger drawings of spiral and wave lines on smartphone screens, aiming to detect early Parkinson's disease.

METHODS: We recruited 58 age-matched participants (28 early idiopathic PD patients: 68.4 ± 5.7 years; 30 healthy controls: 68.0 ± 4.5 years) for two smartphone-based drawing tasks (spiral and wave). A custom-developed app recorded finger touch coordinates, instantaneous movement speed, and timestamps at a sampling frequency of 60 Hz. Our hybrid model combined multi-scale convolutional feature extraction (using parallel 1D-Convolutional branches) with bidirectional temporal pattern recognition (via gated recurrent unit [GRU] networks) to analyze movement abnormalities and detect the disease.

RESULTS: The proposed model demonstrated robust diagnostic performance, achieving a cross-validation accuracy of 87.93% for spiral drawings (89.64% sensitivity, 86.33% specificity). Wave drawings yielded 87.24% accuracy (86.79% sensitivity, 87.67% specificity). The integration of both tasks achieved 91.20% accuracy (95% CI: 89.2%-93.2%) with balanced sensitivity (91.43%) and specificity (91.00%).

CONCLUSION: This study establishes the technical feasibility of a hybrid deep learning framework for early PD detection using smartphone-captured finger motion dynamics. The developed model effectively combines one-dimensional convolutional neural networks with bidirectional GRUs to analyze drawing tasks. Distinct from existing approaches that rely on clinical rating scales, neuroimaging modalities, or stylus-based digital assessments, this telemedicine-compatible method requires only bare-finger interactions on consumer-grade smartphones and enables operator-independent assessments. Furthermore, it facilitates cost-effective and convenient PD assessment in remote healthcare and patient monitoring, particularly in resource-limited settings.

PMID:40658696 | DOI:10.1371/journal.pone.0327733

IEEE Trans Neural Syst Rehabil Eng. 2025 Jul 14;PP. doi: 10.1109/TNSRE.2025.3589089. Online ahead of print.

ABSTRACT

Transcranial ultrasound stimulation is a non-invasive neuromodulation technique characterized by its high spatial resolution and penetration depth, and it has shown an inhibitory effect on epilepsy. However, current applications predominantly employ open-loop transcranial ultrasound stimulation, which lacks the capacity to dynamically respond to seizures. In the present study, we designed and implemented a closed-loop transcranial ultrasound stimulation (CTUS) system comprising a signal acquisition module, a signal preprocessing module, a deep learning network model-based epileptic signal recognition module, and an ultrasound stimulation module to enable real-time detection and ultrasound intervention in the hippocampus of penicillin-induced epileptic mice. The results indicated that the CTUS system could accurately identify epileptic signals, significantly reduce the seizure firing rate, decrease the power intensity and phase-amplitude coupling, and enhance the sample entropy. These findings demonstrated that the deep learning-based CTUS system was efficient in suppressing seizures in mice.

PMID:40658582 | DOI:10.1109/TNSRE.2025.3589089

IEEE Trans Med Imaging. 2025 Jul 14;PP. doi: 10.1109/TMI.2025.3589058. Online ahead of print.

ABSTRACT

The success of deep learning in 3D medical image segmentation hinges on training with a large dataset of fully annotated 3D volumes, which are difficult and time-consuming to acquire. Although recent foundation models (e.g., segment anything model, SAM) can utilize sparse annotations to reduce annotation costs, segmentation tasks involving organs and tissues with blurred boundaries remain challenging. To address this issue, we propose a region uncertainty estimation framework for Computed Tomography (CT) image segmentation using noisy labels. Specifically, we propose a sample-stratified training strategy that stratifies samples according to their varying quality labels, prioritizing confident and fine-grained information at each training stage. This sample-to-voxel level processing enables more reliable supervision information to propagate to noisy label data, thus effectively mitigating the impact of noisy annotations. Moreover, we further design a boundary-guided regional uncertainty estimation module that adapts sample hierarchical training to assist in evaluating sample confidence. Experiments conducted across multiple CT datasets demonstrate the superiority of our proposed method over several competitive approaches under various noise conditions. Our proposed reliable label propagation strategy not only significantly reduces the cost of medical image annotation and robust model training but also improves the segmentation performance in scenarios with imperfect annotations, thus paving the way towards the application of medical segmentation foundation models under low-resource and remote scenarios. Code will be available at https://github.com/KHan-UJS/NoisyLabel.

PMID:40658577 | DOI:10.1109/TMI.2025.3589058

IEEE J Biomed Health Inform. 2025 Jul 14;PP. doi: 10.1109/JBHI.2025.3588122. Online ahead of print.

ABSTRACT

Non-destructive detection methods based on molecular vibrational spectroscopy are pivotal in fields such as analytical chemistry and medical diagnostics. Recent advances have integrated deep learning with vibrational spectroscopy, significantly enhancing spectral recognition accuracy. However, these methods often rely on large annotated spectral datasets, limiting their general applicability. To address this limitation, we propose a novel approach, Global and Local Semantics Mining (GLSM), which leverages self-supervised learning to capture the global and local semantic information of unlabeled spectra, obviating the need for extensive annotated data. We devise two proxy tasks: global semantic mining and local semantic mining. The global semantic mining task is based on the premise that different views of the same spectrum can be mutually transformed, enabling the model to capture domain-invariant features across various perspectives and thereby develop a global understanding of the spectral data. This, in turn, enhances the model's robustness to variations in peak positions. Meanwhile, the local semantic mining task posits that noisy spectra can be reconstructed into noise-free spectra, thereby facilitating the extraction of local patterns and fine-grained details, such as subtle variations in peak intensities. By combining both selfsupervised tasks, our model effectively captures the global and local semantic information of the spectrum. The pretrained model can be fine-tuned with a limited amount of labeled homologous or heterologous spectral data for semi-supervised or transfer learning-based spectral classification. Extensive experiments on three datasets in semisupervised and transfer learning-based spectral recognition tasks comprehensively validate the effectiveness of our GLSM method, demonstrating its significant potential for real-world spectral analysis applications.

PMID:40658574 | DOI:10.1109/JBHI.2025.3588122

IEEE Trans Biomed Eng. 2025 Jul 14;PP. doi: 10.1109/TBME.2025.3588051. Online ahead of print.

ABSTRACT

Wearable biosensors have revolutionized human performance monitoring by enabling real-time assessment of physiological and biomechanical parameters. However, existing solutions lack the ability to simultaneously capture breath-force coordination and muscle activation symmetry in a seamless and non-invasive manner, limiting their applicability in strength training and rehabilitation. This work presents a wearable smart sportswear system that integrates screen-printed graphene-based strain sensors with compact electronics for wireless data transfer and a deep learning framework for real-time classification of exercise execution quality. By leveraging 1D ResNet-18 for feature extraction, the system achieves 92.1% classification accuracy across six exercise conditions, distinguishing between breathing irregularities and asymmetric muscle exertion. Additionally, tSNE analysis and Grad-CAM-based explainability visualization confirm that the network accurately captures biomechanically relevant features, ensuring robust interpretability. The proposed system establishes a foundation for next-generation AI-powered sportswear, with applications in fitness optimization, injury prevention, and adaptive rehabilitation training.

PMID:40658556 | DOI:10.1109/TBME.2025.3588051

Bioinformatics. 2025 Jul 14:btaf397. doi: 10.1093/bioinformatics/btaf397. Online ahead of print.

ABSTRACT

MOTIVATION: DNA methylation plays important roles in various cellular physiological processes in bacteria. Nanopore sequencing has shown the ability to identify different types of DNA methylation from individual bacteria directly. However, existing methods for identifying bacterial methylomes showed inconsistent performances in different methylation motifs in bacteria and didn't fully utilize the different scale information contained in nanopore signals.

RESULTS: We propose a deep-learning method, called Nanoident, for de novo detection of DNA methylation types and methylated base positions in bacteria using Nanopore sequencing. For each targeted motif sequence, Nanoident utilizes five different features, including statistical features extracted from both the nanopore raw signals and the basecalling results of the motif. All the five features are processed by a multi-scale neural network in Nanoident, which extracts information from different receptive fields of the features. The LOOCV (Leave-One-Out Cross Validation) on the dataset containing 7 bacteria samples with 46 methylation motifs shows that, Nanoident achieves ∼10% improvement in accuracy than the previous method. Furthermore, Nanoident achieves ∼13% improvement in accuracy in an independent dataset, which contains 12 methylation motifs. Additionally, we optimize the pipeline for de novo methylation motif enrichment, enabling the discovery of novel methylation motifs.

AVAILABILITY AND IMPLEMENTATION: The source code of Nanoident is freely available at https://github.com/cz-csu/Nanoident and https://doi.org/10.6084/m9.figshare.29252264.

SUPPLEMENTARY INFORMATION: data are available at Bioinformatics online.

PMID:40658463 | DOI:10.1093/bioinformatics/btaf397

Phys Eng Sci Med. 2025 Jul 14. doi: 10.1007/s13246-025-01581-7. Online ahead of print.

ABSTRACT

Accurate liver vessel segmentation is essential for effective liver surgery pre-planning, and reducing surgical risks since it enables the precise localization and extensive assessment of complex vessel structures. Manual liver vessel segmentation is a time-intensive process reliant on operator expertise and skill. The complex, tree-like architecture of hepatic and portal veins, which are interwoven and anatomically variable, further complicates this challenge. This study addresses these challenges by proposing the UNETR (U-Net Transformers) architecture for the multi-class segmentation of portal and hepatic veins in liver CT images. UNETR leverages a transformer-based encoder to effectively capture long-range dependencies, overcoming the limitations of convolutional neural networks (CNNs) in handling complex anatomical structures. The proposed method was evaluated on contrast-enhanced CT images from the IRCAD as well as a locally dataset developed from a hospital. On the local dataset, the UNETR model achieved Dice coefficients of 49.71% for portal veins, 69.39% for hepatic veins, and 76.74% for overall vessel segmentation, while reaching Dice coefficients of 62.54% for vessel segmentation on the IRCAD dataset. These results highlight the method's effectiveness in identifying complex vessel structures across diverse datasets. These findings underscore the critical role of advanced architectures and precise annotations in improving segmentation accuracy. This work provides a foundation for future advancements in automated liver surgery pre-planning, with the potential to enhance clinical outcomes significantly. The implementation code is available on GitHub: https://github.com/saharsarkar/Multiclass-Vessel-Segmentation .

PMID:40658328 | DOI:10.1007/s13246-025-01581-7

Cell Regen. 2025 Jul 14;14(1):32. doi: 10.1186/s13619-025-00246-w.

ABSTRACT

Genetic and microbial factors influence inflammatory bowel disease (IBD), prompting our study on non-invasive biomarkers for enhanced diagnostic precision. Using the XGBoost algorithm and variable analysis and the published metadata, we developed the 10-species signature XGBoost classification model (XGB-IBD10). By using distinct species signatures and prior machine and deep learning models and employing standardization methods to ensure comparability between metagenomic and 16S sequencing data, we constructed classification models to assess the XGB-IBD10 precision and effectiveness. XGB-IBD10 achieved a notable accuracy of 0.8722 in testing samples. In addition, we generated metagenomic sequencing data from collected 181 stool samples to validate our findings, and the model reached an accuracy of 0.8066. The model's performance significantly improved when trained on high-quality data from the Chinese population. Furthermore, the microbiome-based model showed promise in predicting active IBD. Overall, this study identifies promising non-invasive biomarkers associated with IBD, which could greatly enhance diagnostic accuracy.

PMID:40658318 | DOI:10.1186/s13619-025-00246-w

Discov Oncol. 2025 Jul 14;16(1):1330. doi: 10.1007/s12672-025-03119-5.

ABSTRACT

Artificial intelligence (AI) models based on pathological slides have great potential to assist pathologists in disease diagnosis and have become an important research direction in the field of medical image analysis. The aim of this study was to develop an AI model based on whole-slide images to predict the stage of colon cancer. In this study, a total of 100 pathological slides of colon cancer patients were collected as the training set, and 421 pathological slides of colon cancer were downloaded from The Cancer Genome Atlas (TCGA) database as the external validation set. Cellprofiler and CLAM tools were used to extract pathological features, and machine learning algorithms and deep learning algorithms were used to construct prediction models. The area under the curve (AUC) of the best machine learning model was 0.78 in the internal test set and 0.68 in the external test set. The AUC of the deep learning model in the internal test set was 0.889, and the accuracy of the model was 0.854. The AUC of the deep learning model in the external test set was 0.700. The prediction model has the potential to generalize in the process of combining pathological omics diagnosis. Compared with machine learning, deep learning has higher recognition and accuracy of images, and the performance of the model is better.

PMID:40658261 | DOI:10.1007/s12672-025-03119-5

Pediatr Radiol. 2025 Jul 14. doi: 10.1007/s00247-025-06332-0. Online ahead of print.

ABSTRACT

BACKGROUND: The objective and reliable quantification of adenoid size is pivotal for precise clinical diagnosis and the formulation of effective treatment strategies. Conventional manual measurement techniques, however, are often labor-intensive and time-consuming.

OBJECTIVE: To develop and validate a fully automated system for measuring adenoid size using deep learning (DL) on lateral head and neck radiographs.

MATERIALS AND METHODS: In this retrospective study, we analyzed 711 lateral head and neck radiographs collected from two centers between February and July 2023. A DL-based adenoid size measurement system was developed, utilizing Fujioka's method. The system employed the RTMDet network and RTMPose networks for accurate landmark detection, and mathematical formulas were applied to determine adenoid size. To evaluate consistency and reliability of the system, we employed the intra-class correlation coefficient (ICC), mean absolute difference (MAD), and Bland-Altman plots as key assessment metrics.

RESULTS: The DL-based system exhibited high reliability in the prediction of adenoid, nasopharynx, and adenoid-nasopharyngeal ratio measurements, showcasing strong agreement with the reference standard. The results indicated an ICC for adenoid measurements of 0.902 [95%CI, 0.872-0.925], with a MAD of 1.189 and a root mean square (RMS) of 1.974. For nasopharynx measurements, the ICC was 0.868 [95%CI, 0.828-0.899], with a MAD of 1.671 and an RMS of 1.916. Additionally, the adenoid-nasopharyngeal ratio measurements yielded an ICC of 0.911 [95%CI, 0.883-0.932], a MAD of 0.054, and an RMS of 0.076.

CONCLUSIONS: The developed DL-based system effectively automates the measurement of the adenoid-nasopharyngeal ratio, adenoid, and nasopharynx on lateral neck or head radiographs, showcasing high reliability.

PMID:40658209 | DOI:10.1007/s00247-025-06332-0

Spine (Phila Pa 1976). 2025 Jul 14. doi: 10.1097/BRS.0000000000005448. Online ahead of print.

ABSTRACT

STUDY DESIGN: Retrospective study.

OBJECTIVE: To develop a deep learning (DL) model to predict bone cement leakage (BCL) subtypes during percutaneous kyphoplasty (PKP) using preoperative computed tomography (CT) as well as employing multicenter data to evaluate the effectiveness and generalizability of the model.

SUMMARY OF BACKGROUND DATA: DL excels at automatically extracting features from medical images. However, there is a lack of models that can predict BCL subtypes based on preoperative images.

METHODS: This study included an internal dataset for DL model training, validation, and testing as well as an external dataset for additional model testing. Our model integrated a segment localization module based on vertebral segmentation via three-dimensional (3D) U-Net with a classification module based on 3D ResNet-50. Vertebral level mismatch rates were calculated, and confusion matrixes were used to compare the performance of the DL model with that of spine surgeons in predicting BCL subtypes. Furthermore, the simple Cohen's kappa coefficient was used to assess the reliability of spine surgeons and the DL model against the reference standard.

RESULTS: A total of 901 patients containing 997 eligible segments were included in the internal dataset. The model demonstrated a vertebral segment identification accuracy of 96.9%. It also showed high area under the curve (AUC) values of 0.734-0.831 and sensitivities of 0.649-0.900 for BCL prediction in the internal dataset. Similar favorable AUC values of 0.709-0.818 and sensitivities of 0.706-0.857 were observed in the external dataset, indicating the stability and generalizability of the model. Moreover, the model outperformed nonexpert spine surgeons in predicting BCL subtypes, except for type II.

CONCLUSION: The model achieved satisfactory accuracy, reliability, generalizability, and interpretability in predicting BCL subtypes, outperforming nonexpert spine surgeons. This study offers valuable insights for assessing osteoporotic vertebral compression fractures, thereby aiding preoperative surgical decision-making.

LEVEL OF EVIDENCE: 3.

PMID:40658115 | DOI:10.1097/BRS.0000000000005448

Elife. 2025 Jul 14;13:RP104901. doi: 10.7554/eLife.104901.

ABSTRACT

To design safe, selective, and effective new therapies, there must be a deep understanding of the structure and function of the drug target. One of the most difficult problems to solve has been the resolution of discrete conformational states of transmembrane ion channel proteins. An example is KV11.1 (hERG), comprising the primary cardiac repolarizing current, Ikr. hERG is a notorious drug anti-target against which all promising drugs are screened to determine potential for arrhythmia. Drug interactions with the hERG inactivated state are linked to elevated arrhythmia risk, and drugs may become trapped during channel closure. While prior studies have applied AlphaFold to predict alternative protein conformations, we show that the inclusion of carefully chosen structural templates can guide these predictions toward distinct functional states. This targeted modeling approach is validated through comparisons with experimental data, including proposed state-dependent structural features, drug interactions from molecular docking, and ion conduction properties from molecular dynamics simulations. Remarkably, AlphaFold not only predicts inactivation mechanisms of the hERG channel that prevent ion conduction but also uncovers novel molecular features explaining enhanced drug binding observed during inactivation, offering a deeper understanding of hERG channel function and pharmacology. Furthermore, leveraging AlphaFold-derived states enhances computational screening by significantly improving agreement with experimental drug affinities, an important advance for hERG as a key drug safety target where traditional single-state models miss critical state-dependent effects. By mapping protein residue interaction networks across closed, open, and inactivated states, we identified critical residues driving state transitions validated by prior mutagenesis studies. This innovative methodology sets a new benchmark for integrating deep learning-based protein structure prediction with experimental validation. It also offers a broadly applicable approach using AlphaFold to predict discrete protein conformations, reconcile disparate data, and uncover novel structure-function relationships, ultimately advancing drug safety screening and enabling the design of safer therapeutics.

PMID:40658102 | DOI:10.7554/eLife.104901

J Acoust Soc Am. 2025 Jul 1;158(1):380-390. doi: 10.1121/10.0037197.

ABSTRACT

Low latency is an essential requirement for noise reduction in real-world devices such as hearing aids and cochlear implants. Reducing the algorithmic latency of a deep neural network charged with noise reduction allows additional time for other processing. However, a larger analysis window may be advantageous to the performance of the network. This trade-off is currently examined with regard to human speech-intelligibility performance. The algorithmic latency of the attentive recurrent network (ARN) was modified by reducing the size of the analysis time frame. The ARN model was talker, noise, and recording-channel independent, and fully causal. Listeners with hearing loss and with normal hearing heard sentences in babble at various signal-to-noise ratios. Large increases in intelligibility were observed as a result of noise reduction, especially for the listeners with hearing loss and at less favorable signal-to-noise ratios. Slightly larger objective measures of network performance were observed at larger latencies. But more critically, human performance was essentially unchanged as algorithmic latency was reduced from 20 to 10 or 5 ms. These results are discussed in the context of overall design and implementation of deep-learning based noise reduction, and information on latency requirements for human listeners is summarized.

PMID:40657924 | DOI:10.1121/10.0037197