Acta Ophthalmol. 2025 May 21. doi: 10.1111/aos.17518. Online ahead of print.

ABSTRACT

PURPOSE: Accurate assessment of cystoid macular oedema (CMO) in patients with retinitis pigmentosa (RP) on spectral-domain optical coherence tomography (SD-OCT) is crucial for tracking disease progression and may serve as a therapeutic endpoint. Manual CMO segmentation is labour-intensive and prone to variability, making artificial intelligence (AI) an appealing solution to improve accuracy and efficiency. This study aimed to validate a deep learning (DL) model for automated CMO detection and quantification on SD-OCT scans in patients with RP.

METHODS: A segmentation model based on the no-new-Unet (nnU-Net) architecture was trained on 112 OCT volumes from the RETOUCH dataset (70 for training, 42 for validation). The model was externally tested on 37 SD-OCT scans from RP patients, with annotations from three expert graders. Performance was assessed using the Dice similarity coefficient and intraclass correlation coefficient (ICC).

RESULTS: For randomly selected central B-scans, the model achieved a mean Dice score of 0.889 ± 0.002 standard deviation (SD), while observers scored 0.878 ± 0.007 SD. The ICC for the model was 0.945 ± 0.014 SD, compared to 0.979 ± 0.008 SD for observers. On manually chosen central B-scans, Dice scores were 0.936 ± 0.005 SD for the model and 0.946 ± 0.012 SD for observers, with ICC values of 0.964 ± 0.011 SD and 0.981 ± 0.011 SD, respectively.

CONCLUSIONS: This DL model reliably segments CMO in RP, achieving performance comparable to human graders. It can enhance the efficiency and precision of CMO quantification, reducing variability in clinical practice and trials.

PMID:40396533 | DOI:10.1111/aos.17518

NMR Biomed. 2025 Jul;38(7):e70068. doi: 10.1002/nbm.70068.

ABSTRACT

Chemical exchange saturation transfer (CEST) MRI is a powerful molecular imaging technique for detecting metabolites through proton exchange. While CEST MRI provides high sensitivity, its clinical application is hindered by prolonged scan time due to the need for imaging across numerous frequency offsets for parameter estimation. Since scan time is directly proportional to the number of frequency offsets, identifying and selecting the most informative frequency can significantly reduce acquisition time. We propose a novel deep learning-based framework that integrates frequency selection and parameter estimation to accelerate CEST MRI. Our method leverages channel pruning via batch normalization to identify the most informative frequency offsets while simultaneously training the network for accurate parametric map prediction. Using data from six healthy volunteers, channel pruning selects 13 informative frequency offsets out of 53 without compromising map quality. Images from selected frequency offsets were reconstructed using the MR Multitasking method, which employs a low-rank tensor model to enable under-sampling of k-space lines for each frequency offset, further reducing scan time. Predicted parametric maps of amide proton transfer (APT), nuclear overhauser effect (NOE), and magnetization transfer (MT) based on these selected frequencies were comparable in quality to maps generated using all frequency offsets, achieving superior performance compared to Fisher information-based selection methods from our previous work. This integrated approach has the potential to reduce the whole-brain CEST MRI scan time from the original 5:30 min to under 1:30 min without compromising map quality. By leveraging deep learning for frequency selection and parametric map prediction, the proposed framework demonstrates its potential for efficient and practical clinical implementation. Future studies will focus on extending this method to patient populations and addressing challenges such as B0 inhomogeneity and abnormal signal variation in diseased tissues.

PMID:40396230 | DOI:10.1002/nbm.70068

J Clin Ultrasound. 2025 May 21. doi: 10.1002/jcu.24058. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: Our objective is to develop and validate a deep learning radiomics nomogram (DLRN) based on preoperative ultrasound images and clinical features, for predicting the malignancy of thyroid nodules with indeterminate cytology (Bethesda III).

MATERIALS AND METHODS: Between June 2017 and June 2022, we conducted a retrospective study on 194 patients with surgically confirmed indeterminate cytology (Bethesda III) in our hospital. The training and internal validation cohorts were comprised of 155 and 39 patients, in a 7:3 ratio. To facilitate external validation, we selected an additional 80 patients from each of the remaining two medical centers. Utilizing preoperative ultrasound data, we obtained imaging markers that encompass both deep learning and manually radiomic features. After feature selection, we developed a comprehensive diagnostic model to evaluate the predictive value for Bethesda III benign and malignant cases. The model's diagnostic accuracy, calibration, and clinical applicability were systematically assessed.

RESULTS: The results showed that the prediction model, which integrated 512 DTL features extracted from the pre-trained Resnet34 network, ultrasound radiomics, and clinical features, exhibited superior stability in distinguishing between benign and malignant indeterminate thyroid nodules (Bethesda Class III). In the validation set, the AUC was 0.92 (95% CI: 0.831-1.000), and the accuracy, sensitivity, specificity, precision, and recall were 0.897, 0.882, 0.909, 0.882, and 0.882, respectively.

CONCLUSION: The comprehensive multidimensional data model based on deep transfer learning, ultrasound radiomics features, and clinical characteristics can effectively distinguish the benign and malignant indeterminate thyroid nodules (Bethesda Class III), providing valuable guidance for treatment selection in patients with indeterminate thyroid nodules (Bethesda Class III).

PMID:40396203 | DOI:10.1002/jcu.24058

Clin Ophthalmol. 2025 May 16;19:1593-1607. doi: 10.2147/OPTH.S521558. eCollection 2025.

ABSTRACT

BACKGROUND: Idiopathic macular hole is an ophthalmic disease that seriously affects vision, and its early diagnosis and treatment have important clinical significance to reduce the occurrence of blindness. At present, OCT is the gold standard for diagnosing this disease, but its application is limited due to the need for professional ophthalmologist to diagnose it. The introduction of artificial intelligence will break this situation and make its diagnosis efficient, and how to build an effective predictive model is the key to the problem, and more clinical trials are still needed to verify it.

OBJECTIVE: This study aims to evaluate the role of deep learning systems in Idiopathic Macular Hole diagnosis, grading, and prediction.

METHODS: A single-center, retrospective study used binocular OCT images from IMH patients at the First Affiliated Hospital of Nanchang University (November 2019 - January 2023). A deep learning algorithm, including traditional omics, Resnet101, and fusion models incorporating multi-feature fusion and transfer learning, was developed. Model performance was evaluated using accuracy and AUC. Logistic regression analyzed clinical factors, and a nomogram predicted surgical risk. Analysis was conducted with SPSS 22.0 and R 3.6.3. P < 0.05 was statistically significant.

RESULTS: Among 229 OCT images, the traditional omics, Resnet101, and fusion models achieved accuracies of 93%, 94%, and 95%, respectively, in the training set. In the test set, the fusion model and Resnet101 correctly identified 39 images, while the traditional omics model identified 35. The nomogram had a C-index of 0.996, with macular hole diameter most strongly associated with surgical risk.

CONCLUSION: The deep learning system with transfer learning and multi-feature fusion effectively diagnoses and grades IMH from OCT images.

PMID:40396157 | PMC:PMC12091069 | DOI:10.2147/OPTH.S521558

J Hand Microsurg. 2025 Apr 22;17(4):100257. doi: 10.1016/j.jham.2025.100257. eCollection 2025 Jul.

ABSTRACT

PURPOSE: Artificial intelligence (AI) is transforming orthopedics by improving diagnostic accuracy, optimizing surgical planning, and personalizing treatment approaches. This review evaluates the applications of AI in orthopedics, focusing on its impact on patient care, the efficacy of AI methodologies, and challenges in integrating these technologies into clinical practice.

METHODS: A comprehensive literature search was conducted across PubMed, Scopus, and Google Scholar for articles published up to 28th February 2025. Inclusion criteria included studies addressing AI applications in orthopedics, while non-peer-reviewed and non-English publications were excluded. Data extraction focused on AI technologies, applications, outcomes, and the advantages or limitations of AI integration.

RESULTS: Findings demonstrate AI's effectiveness in areas such as fracture detection and treatment planning, mainly through machine learning and deep learning. AI has improved outcomes in joint reconstruction, spine surgery, and rehabilitation. However, challenges such as data standardization and clinical validation remain.

CONCLUSION: The review highlights AI's potential to revolutionize orthopedic practice, emphasizing the need for ongoing research to overcome barriers to adoption. Future directions should prioritize multi-center clinical trials, enhanced data protocols, and stakeholder collaboration to ensure ethical and effective AI implementation, ultimately improving patient outcomes and care delivery.

PMID:40395968 | PMC:PMC12088811 | DOI:10.1016/j.jham.2025.100257

ArXiv [Preprint]. 2025 Apr 29:arXiv:2504.20405v1.

ABSTRACT

While deep learning has shown strong performance in musculoskeletal imaging, existing work has largely focused on pathologies where diagnosis is not a clinical challenge, leaving more difficult problems underexplored, such as detecting Bankart lesions (anterior-inferior glenoid labral tears) on standard MRIs. Diagnosing these lesions is challenging due to their subtle imaging features, often leading to reliance on invasive MRI arthrograms (MRAs). This study introduces ScopeMRI, the first publicly available, expert-annotated dataset for shoulder pathologies, and presents a deep learning (DL) framework for detecting Bankart lesions on both standard MRIs and MRAs. ScopeMRI includes 586 shoulder MRIs (335 standard, 251 MRAs) from 558 patients who underwent arthroscopy. Ground truth labels were derived from intraoperative findings, the gold standard for diagnosis. Separate DL models for MRAs and standard MRIs were trained using a combination of CNNs and transformers. Predictions from sagittal, axial, and coronal views were ensembled to optimize performance. The models were evaluated on a 20% hold-out test set (117 MRIs: 46 MRAs, 71 standard MRIs). The models achieved an AUC of 0.91 and 0.93, sensitivity of 83% and 94%, and specificity of 91% and 86% for standard MRIs and MRAs, respectively. Notably, model performance on non-invasive standard MRIs matched or surpassed radiologists interpreting MRAs. External validation demonstrated initial generalizability across imaging protocols. This study demonstrates that DL models can achieve radiologist-level diagnostic performance on standard MRIs, reducing the need for invasive MRAs. By releasing ScopeMRI and a modular codebase for training and evaluating deep learning models on 3D medical imaging data, we aim to accelerate research in musculoskeletal imaging and support the development of new datasets for clinically challenging diagnostic tasks.

PMID:40395941 | PMC:PMC12091705

ArXiv [Preprint]. 2025 May 17:arXiv:2504.08713v3.

ABSTRACT

Deep learning-based electrocardiogram (ECG) classification has shown impressive performance but clinical adoption has been slowed by the lack of transparent and faithful explanations. Post hoc methods such as saliency maps may fail to reflect a model's true decision process. Prototype-based reasoning offers a more transparent alternative by grounding decisions in similarity to learned representations of real ECG segments, enabling faithful, case-based explanations. We introduce ProtoECGNet, a prototype-based deep learning model for interpretable, multi-label ECG classification. ProtoECGNet employs a structured, multi-branch architecture that reflects clinical interpretation workflows: it integrates a 1D CNN with global prototypes for rhythm classification, a 2D CNN with time-localized prototypes for morphology-based reasoning, and a 2D CNN with global prototypes for diffuse abnormalities. Each branch is trained with a prototype loss designed for multi-label learning, combining clustering, separation, diversity, and a novel contrastive loss that encourages appropriate separation between prototypes of unrelated classes while allowing clustering for frequently co-occurring diagnoses. We evaluate ProtoECGNet on all 71 diagnostic labels from the PTB-XL dataset, demonstrating competitive performance relative to state-of-the-art black-box models while providing structured, case-based explanations. To assess prototype quality, we conduct a structured clinician review of the final model's projected prototypes, finding that they are rated as representative and clear. ProtoECGNet shows that prototype learning can be effectively scaled to complex, multi-label time-series classification, offering a practical path toward transparent and trustworthy deep learning models for clinical decision support.

PMID:40395940 | PMC:PMC12091707

MethodsX. 2025 Apr 28;14:103337. doi: 10.1016/j.mex.2025.103337. eCollection 2025 Jun.

ABSTRACT

This paper proposes a model of a doubly clamped electrically actuated microbeam, a structure frequently utilized in microelectromechanical systems (MEMS). The model investigates the effect of electrostatic forces on the deflection of the beam, based on the Euler-Bernoulli beam theory. The Galerkin technique is employed to calculate the beam's deflection, while the parallel plate capacitor model simulates the electric field. We also evaluate the performance of multi-walled carbon nanotubes (MWCNTs) in MEMS. MWCNTs are promising for MEMS applications due to their significant thermal, mechanical, and electrical properties. However, predicting the behavior of these systems is challenging due to their stiffness, parametric sensitivity, and non-linearity. Deep learning strategies for handling dynamical systems are a rapidly emerging field of research. In this paper, we employ a machine learning method, called deep neural networks (DNN), to solve the non-linear systems that arise in MEMS. The primary aim of this study is to investigate the nonlinear vibration properties of MEMS oscillators, specifically those related to nanotubes and electrically actuated microbeams, using DNN algorithms. Different optimizers are used to analyze the performance and capability of these non-linear dynamical models. Numerical simulations and graphical demonstrations are presented to verify the accuracy and efficiency of the algorithm.•The study develops a novel DNN-based model to solve non-linear systems in MEMS, particularly for oscillators with MWCNTs.•Deep learning optimizers are applied to improve the accuracy and efficiency of predicting MEMS behavior.•Numerical simulations confirm the effectiveness of the proposed methodology.

PMID:40395931 | PMC:PMC12090320 | DOI:10.1016/j.mex.2025.103337

MethodsX. 2025 Apr 25;14:103333. doi: 10.1016/j.mex.2025.103333. eCollection 2025 Jun.

ABSTRACT

Parkinson's disease (PD) is a neurodegenerative condition that progressively affects motor function and causes tremors, rigidity, and bradykinesia. Detection of PD at an early stage is important to ensure timely intervention and better patient outcomes. This study uses deep learning algorithms to classify spiral images traced by patients as an inexpensive diagnostic technique for the detection of PD. A database consists of spiral images drawn manually by PD patients and normal individuals, divided into training and testing sets. To discriminate between spiral drawings of Parkinsonian and healthy cases four Convolutional Neural Network (CNN) architecture like DenseNet121, InceptionV3, VGG16, and LeNet are used. Followed by transfer learning which is employed to improve model performance by extracting fine motor impairment patterns in the spirals. DenseNet121 and InceptionV3 achieve competitive performance with 98.44 % accuracy, whereas VGG16 demonstrates excellent feature extraction performance. The study emphasizes the relevance of deep learning in non-invasive PD diagnosis, as a consistent, efficient, and automated method of early detection. The future can be directed towards the combination of spiral images with other biomarkers or a broader data set with other motor measures in a wider disease assessment.•The study focuses on enhancing features extraction by leveraging hybrid deep learning models, improving classification performance.•Implementation of features scaling leads to better model performance, with improved accuracy.•The comparative analysis of CNN architecture provides valuable insights into balancing computational efficiency and classification performance.

PMID:40395929 | PMC:PMC12090310 | DOI:10.1016/j.mex.2025.103333

Front Hum Neurosci. 2025 May 6;19:1583342. doi: 10.3389/fnhum.2025.1583342. eCollection 2025.

ABSTRACT

INTRODUCTION: Electroencephalography (EEG) signals frequently contain substantial noise and interference, which can obscure clinically and scientifically relevant features. Traditional denoising approaches, such as linear filtering or wavelet thresholding, often struggle with nonlinear or time-varying artifacts. In response, the present study explores a Generative Adversarial Network (GAN) framework to enhance EEG signal quality, focusing on two variants: a conventional GAN model and a Wasserstein GAN with Gradient Penalty (WGAN-GP).

METHODS: Data were obtained from two distinct EEG datasets: a "healthy" set of 64-channel recordings collected during various motor/imagery tasks, and an "unhealthy" set of 18-channel recordings from individuals with orthopedic impairments. Both datasets underwent comprehensive preprocessing, including band-pass filtering (8-30 Hz), channel standardization, and artifact trimming. The training stage involved adversarial learning, in which a generator sought to reconstruct clean EEG signals while a discriminator (or critic in the case of WGAN-GP) attempted to distinguish between real and generated signals. The model evaluation was conducted using quantitative metrics such as signal-to-noise ratio (SNR), peak signal-to-noise ratio (PSNR), correlation coefficient, mutual information, and dynamic time warping (DTW) distance.

RESULTS: Experimental findings indicate that adversarial learning substantially improves EEG signal fidelity across multiple quantitative metrics. Specifically, WGAN-GP achieved an SNR of up to 14.47 dB (compared to 12.37 dB for the standard GAN) and exhibited greater training stability, as evidenced by consistently lower relative root mean squared error (RRMSE) values. In contrast, the conventional GAN model excelled in preserving finer signal details, reflected in a PSNR of 19.28 dB and a correlation coefficient exceeding 0.90 in several recordings. Both adversarial frameworks outperformed classical wavelet-based thresholding and linear filtering methods, demonstrating superior adaptability to nonlinear distortions and dynamic interference patterns in EEG time-series data.

DISCUSSION: By systematically comparing standard GAN and WGAN-GP architectures, this study highlights a practical trade-off between aggressive noise suppression and high-fidelity signal reconstruction. The demonstrated improvements in signal quality underscore the promise of adversarially trained models for applications ranging from basic neuroscience research to real-time brain-computer interfaces (BCIs) in clinical or consumer-grade settings. The results further suggest that GAN-based frameworks can be easily scaled to next-generation wireless networks and complex electrophysiological datasets, offering robust and dynamic solutions to long-standing challenges in EEG denoising.

PMID:40395688 | PMC:PMC12089060 | DOI:10.3389/fnhum.2025.1583342

Front Bioeng Biotechnol. 2025 May 6;13:1590962. doi: 10.3389/fbioe.2025.1590962. eCollection 2025.

ABSTRACT

BACKGROUND: The application of deep learning techniques in medical image analysis has shown great potential in assisting clinical diagnosis. This study focuses on the development and evaluation of deep learning models for the classification of knee joint injuries using Magnetic Resonance Imaging (MRI) data. The research aims to provide an efficient and reliable tool for clinicians to aid in the diagnosis of knee joint disorders, particularly focusing on Anterior Cruciate Ligament (ACL) tears.

METHODS: KneeXNet leverages the power of graph convolutional networks (GCNs) to capture the intricate spatial dependencies and hierarchical features in knee MRI scans. The proposed model consists of three main components: a graph construction module, graph convolutional layers, and a multi-scale feature fusion module. Additionally, a contrastive learning scheme is employed to enhance the model's discriminative power and robustness. The MRNet dataset, consisting of knee MRI scans from 1,370 patients, is used to train and validate KneeXNet.

RESULTS: The performance of KneeXNet is evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC) metric and compared to state-of-the-art methods, including traditional machine learning approaches and deep learning models. KneeXNet consistently outperforms the competing methods, achieving AUC scores of 0.985, 0.972, and 0.968 for the detection of knee joint abnormalities, ACL tears, and meniscal tears, respectively. The cross-dataset evaluation further validates the generalization ability of KneeXNet, maintaining its superior performance on an independent dataset.

APPLICATION: To facilitate the clinical application of KneeXNet, a user-friendly web interface is developed using the Django framework. This interface allows users to upload MRI scans, view diagnostic results, and interact with the system seamlessly. The integration of Grad-CAM visualizations enhances the interpretability of KneeXNet, enabling radiologists to understand and validate the model's decision-making process.

PMID:40395675 | PMC:PMC12088959 | DOI:10.3389/fbioe.2025.1590962

Se Pu. 2025 Jun;43(6):696-704. doi: 10.3724/SP.J.1123.2024.05012.

ABSTRACT

Gel electrophoresis is used to separate and analyze macromolecules （such as DNA， RNA， and proteins） and their fragments， and highly reproducible and efficient automatic band-detection methods have been developed to analyze gel images. Uneven background， low contrast， lane distortion， blurred band edges， and geometric deformation pose detection-accuracy challenges during automatic band detection. In order to address these issues， various correction algorithms have been proposed； however， these algorithms rely on researcher experience to adjust and optimize parameters based on image characteristics， which introduces human error while qualitatively and quantitatively processing bands. Isoelectric focusing （IEF） gel electrophoresis separates proteins with high-resolution based on isoelectric point （pI） differences. Microarray IEF （mIEF） is used for the auxiliary diagnosis of diabetes and adult β-thalassemia owing to operational ease， low sample consumption， and high throughout. This diagnostic method relies on accurately positioning and precisely determining protein bands. To avoid errors associated with correction algorithms during band analysis， this paper introduces a method for rapidly recognizing bands in gel electrophoresis patterns that relies on a deep learning object detection algorithm， and uses it to quantify and classify the IEF electrophoresis pattern of hemoglobin （Hb）. We used mIEF experiments to collect 1 665 pI-marker-free Hb IEF images as a model dataset to train the YOLOv8 model. The trained model accepts a Hb IEF image as input and infers band bounding boxes and classification results. Using inference data， the gray intensities of the pixels in each band area are summed to determine the content of each protein. The background and foreground of the image need to be separated prior to summing the abovementioned gray intensities， and the threshold method is used to achieve this. The threshold is defined as the average intensity of the background area， which is obtained by summing and averaging the background intensities of gel areas between the detection bounding boxes of each protein band. The baseline band areas are unified after removing the background. This method only requires the input image， directly outputs the corresponding electrophoretic band information， and does not rely on the experience of professionals nor is it affected by factors such as lane distortion or band deformation. In addition， the developed method does not depend on pI markers for qualitatively determining bands， thereby reducing experimental costs and improving detection efficiency. YOLOv8n delivered a detection accuracy of 92.9% and an inference time of 0.6 ms while using limited computing resources. Using Hb A2 as an example， we compared its content measured using the developed method with clinical data. The quantitative results were subjected to regression analysis， which delivered a linearity of 0.981 2 and a correlation coefficient of 0.980 0. We also used the Bland-Altman analysis method to verify that these two values are highly consistent. Compared with the traditional automatic band detection methods， the method developed in this study is fast， accurate， more repeatable， and stable， and can be used to determine the Hb A2 content in clinical practice， thereby potentially assisting in the auxiliary diagnosis of adult β-thalassemia.

PMID:40394749 | DOI:10.3724/SP.J.1123.2024.05012

BMC Musculoskelet Disord. 2025 May 20;26(1):498. doi: 10.1186/s12891-025-08733-6.

ABSTRACT

OBJECTIVE: The aim of this study was to develop a hybrid diagnostic framework integrating radiomic and deep features for accurate and reproducible detection and classification of wrist fractures using X-ray images.

MATERIALS AND METHODS: A total of 3,537 X-ray images, including 1,871 fracture and 1,666 non-fracture cases, were collected from three healthcare centers. Radiomic features were extracted using the PyRadiomics library, and deep features were derived from the bottleneck layer of an autoencoder. Both feature modalities underwent reliability assessment via Intraclass Correlation Coefficient (ICC) and cosine similarity. Feature selection methods, including ANOVA, Mutual Information (MI), Principal Component Analysis (PCA), and Recursive Feature Elimination (RFE), were applied to optimize the feature set. Classifiers such as XGBoost, CatBoost, Random Forest, and a Voting Classifier were used to evaluate diagnostic performance. The dataset was divided into training (70%) and testing (30%) sets, and metrics such as accuracy, sensitivity, and AUC-ROC were used for evaluation.

RESULTS: The combined radiomic and deep feature approach consistently outperformed standalone methods. The Voting Classifier paired with MI achieved the highest performance, with a test accuracy of 95%, sensitivity of 94%, and AUC-ROC of 96%. The end-to-end model achieved competitive results with an accuracy of 93% and AUC-ROC of 94%. SHAP analysis and t-SNE visualizations confirmed the interpretability and robustness of the selected features.

CONCLUSIONS: This hybrid framework demonstrates the potential for integrating radiomic and deep features to enhance diagnostic performance for wrist and forearm fractures, providing a reliable and interpretable solution suitable for clinical applications.

PMID:40394557 | DOI:10.1186/s12891-025-08733-6

BMC Bioinformatics. 2025 May 20;26(1):130. doi: 10.1186/s12859-025-06138-9.

ABSTRACT

Single-cell RNA sequencing (scRNA-seq) enables high-resolution analysis of cellular heterogeneity, but dropout events, where gene expression is undetected in individual cells, present a significant challenge. We propose scMASKGAN, which transforms matrix imputation into a pixel restoration task to improve the recovery of missing gene expression data. Specifically, we integrate masking, convolutional neural networks (CNNs), attention mechanisms, and residual networks (ResNets) to effectively address dropout events in scRNA-seq data. The masking mechanism ensures the preservation of complete cellular information, while convolution and attention mechanisms are employed to capture both global and local features. Residual networks augment feature representation and effectively mitigate the risk of model overfitting. Additionally, cell-type labels are incorporated as constraints to guide the methods in learning more accurate cellular features. Finally, multiple experiments were conducted to evaluate the methods' performance using seven different data types and scRNA-seq data from ten neuroblastoma samples. The results demonstrate that the data imputed by scMASKGAN not only perform excellently across various evaluation metrics but also significantly enhance the effectiveness of downstream analyses, enabling a more comprehensive exploration of underlying biological information.

PMID:40394489 | DOI:10.1186/s12859-025-06138-9

J Imaging Inform Med. 2025 May 20. doi: 10.1007/s10278-025-01539-x. Online ahead of print.

ABSTRACT

The objective of this study is to enhance the understanding of ophthalmic disease physiology and genetic architecture through the analysis of optical coherence tomography (OCT) images using artificial intelligence (AI). We introduce a novel AI methodology that addresses the challenge of transferring OCT phenotypes across datasets. The approach employs unsupervised and self-supervised learning techniques to phenotype and cluster OCT-derived retinal layer thicknesses, using glaucoma as a model disease. Our method integrates deep learning, manifold learning, and a Gaussian mixture model to identify distinct phenotypic clusters. Across two large datasets-Massachusetts Eye and Ear (MEE; 18,985 images) and UK Biobank (UKBB; 86,115 images)-the model identified 9 to 11 phenotypic clusters per retinal layer, which were clinically meaningful and showed consistent patterns across datasets. Pearson correlation analysis confirmed the intra-cluster similarity, with within-cluster correlations exceeding inter-cluster correlations (Supplemental Figs. 4-5). Clinical associations showed that specific phenotypes correlated strongly with glaucoma severity markers, including visual field mean deviation (e.g., 12.57±10.1 for phenotype 6) and cup-to-disc ratio (e.g., 0.694±0.237). These results validate the robustness of the model and its ability to generalize across datasets. This work advances OCT-based phenotyping, enabling phenotype transfer and facilitating translational research in disease mechanisms and genetic discovery.

PMID:40394321 | DOI:10.1007/s10278-025-01539-x

J Imaging Inform Med. 2025 May 20. doi: 10.1007/s10278-025-01543-1. Online ahead of print.

ABSTRACT

Prostate cancer is the most prevalent solid tumor in males and one of the most common causes of male mortality. It is the most common type of cancer in men, a major global public health issue, and accounts for up to 7.3% of all male cancer diagnoses worldwide. To optimize patient outcomes and ensure therapeutic success, an accurate diagnosis must be made promptly. To achieve this, we focused on using ResNet50, a convolutional neural network (CNN) architecture, to analyze prostate histological images to classify prostate cancer. ResNet50, due to its efficiency in medical image classification, was used to classify the histological images as benign or malignant. In this study, a total of 1276 prostate biopsy images were used on the ResNet50 model. We employed evaluation metrics such as accuracy, precision, recall, and F1 score. The results showed that the ResNet50 model performed excellently with an overall accuracy of 0.98, 1.00 as precision, 0.98 as recall, and 0.97 as F1 score for benign. The malignant histological image has 0.99, 0.98, and 0.97 as precision, recall, and F1 scores. It also recorded a 95% confidence interval (CI) for accuracy as (0.91, 1.00) and a performance gain of 4.26% compared to MobileNet and CNN-RNN. The result of our model was also compared with the state-of-the-art (SOTA) DL models to ensure robustness. This study has demonstrated the potential of the ResNet50 model in the classification of prostate cancer. Again, the clinical integration of the results of this study will aid decision-makers in enhancing patient outcomes.

PMID:40394318 | DOI:10.1007/s10278-025-01543-1

Eur Arch Otorhinolaryngol. 2025 May 20. doi: 10.1007/s00405-025-09451-4. Online ahead of print.

ABSTRACT

PURPOSE: Deep learning, a subset of machine learning, is widely utilized in medical applications. Identifying maxillary sinus pathologies before surgical interventions is crucial for ensuring successful treatment outcomes. Cone beam computed tomography (CBCT) is commonly employed for maxillary sinus evaluations due to its high resolution and lower radiation exposure. This study aims to assess the accuracy of artificial intelligence (AI) algorithms in detecting maxillary sinus pathologies from CBCT scans.

METHODS: A dataset comprising 1000 maxillary sinuses (MS) from 500 patients was analyzed using CBCT. Sinuses were categorized based on the presence or absence of pathology, followed by segmentation of the maxillary sinus. Manual segmentation masks were generated using the semiautomatic software ITK-SNAP, which served as a reference for comparison. A convolutional neural network (CNN)-based machine learning model was then implemented to automatically segment maxillary sinus pathologies from CBCT images. To evaluate segmentation accuracy, metrics such as the Dice similarity coefficient (DSC) and intersection over union (IoU) were utilized by comparing AI-generated results with human-generated segmentations.

RESULTS: The automated segmentation model achieved a Dice score of 0.923, a recall of 0.979, an IoU of 0.887, an F1 score of 0.970, and a precision of 0.963.

CONCLUSION: This study successfully developed an AI-driven approach for segmenting maxillary sinus pathologies in CBCT images. The findings highlight the potential of this method for rapid and accurate clinical assessment of maxillary sinus conditions using CBCT imaging.

PMID:40394252 | DOI:10.1007/s00405-025-09451-4

Sci Rep. 2025 May 20;15(1):17481. doi: 10.1038/s41598-025-01763-z.

ABSTRACT

Modeling the 3D structures of cells and tissues is crucial in biology. Sequential cross-sectional images from electron microscopy provide high-resolution intracellular structure information. The segmentation of complex cell structures remains a laborious manual task for experts, demanding time and effort. This bottleneck in analyzing biological images requires efficient and automated solutions. In this study, the deep learning-based automated segmentation of biological images was explored to enable accurate reconstruction of the 3D structures of cells and organelles. An analysis system for the cell images of Cyanidioschyzon merolae, a primitive unicellular red algae, was constructed. This system utilizes sequential cross-sectional images captured by a focused ion beam scanning electron microscope (FIB-SEM). A U-Net was adopted and training was performed to identify and segment cell organelles from single-cell images. In addition, the segment anything model (SAM) and 3D watershed algorithm were employed to extract individual 3D images of each cell from large-scale microscope images containing numerous cells. Finally, the trained U-Net was applied to segment each structure within these 3D images. Through this procedure, the creation of 3D cell models could be fully automated. The adoption of other deep learning techniques and combinations of image processing methods will also be explored to enhance the segmentation accuracy further.

PMID:40394179 | DOI:10.1038/s41598-025-01763-z

Sci Rep. 2025 May 20;15(1):17507. doi: 10.1038/s41598-025-01999-9.

ABSTRACT

Enterprise knowledge retrieval faces challenges like sparse data and inefficient cross-domain knowledge transfer, hindering traditional methods. To address this, we develop a cross-domain recommendation model (CDR-VAE), combining a hybrid autoencoder with domain alignment, and test its effectiveness on an enterprise dataset and the Movies&Books benchmark. At a top-5 recommendation length, CDR-VAE scores HR = 0.642, Recall = 0.432, NDCG = 0.715, outperforming existing models. Removing shared latent representations reduces HR to 0.701, proving their necessity for cross-domain learning. In enterprise applications, high-activity users favor technical reports (0.903), while low-activity users shift toward cross-domain content like industry standards (0.701), confirming the model's robustness in sparse scenarios. CDR-VAE successfully tackles sparsity and cross-domain barriers, advancing enterprise knowledge management. This work provides theoretical and practical insights for deep learning-based recommendation systems in data-scarce environments.

PMID:40394166 | DOI:10.1038/s41598-025-01999-9

Sci Rep. 2025 May 20;15(1):17516. doi: 10.1038/s41598-025-02152-2.

ABSTRACT

The swift development of the Internet of Things (IoT) devices has created a pressing need for effective cybersecurity measures. They are vulnerable to different cyber threats that can compromise the functionality and security of urban systems. Distributed Denial of Service (DDoS) attacks are among IoT networks' most challenging and destructive cyber threats. With the rapid growth in IoT devices and users, the vulnerability of IoT devices to such attacks has enhanced significantly, making DDoS attacks a predominant threat. This work introduces several approaches for effectively detecting IoT-based DDoS threats. Classical machine learning (ML) techniques mostly face difficulty in managing real-world traffic characteristics effectually, making them less appropriate for detecting DDoS attacks. In contrast, Artificial Intelligence (AI)-based methods have proven more effective in detecting cyber-attacks than conventional approaches. This manuscript proposes an effective Feature Pruning with Optimal Deep Learning-based DDoS Attack Detection (FPODL-DDoSAD) technique in the IoT framework. The FPODL-DDoSAD technique initially uses a min-max scalar for the data scaling into the standard layout. Besides, the feature pruning process is performed using an improved pelican optimization algorithm (IPOA), which enables the choice of an optimal subset of features. Meanwhile, DDoS attacks are recognized using a sparse denoising autoencoder (SDAE) model. Furthermore, the parameter tuning of the SDAE classifier is accomplished by utilizing the Fish Migration Optimizer (FMO) technique. The experimental values of the FPODL-DDoSAD approach are assessed on the benchmark BoT-IoT dataset. The comparison study of the FPODL-DDoSAD method demonstrates a superior accuracy value of 99.80% over existing techniques.

PMID:40394115 | DOI:10.1038/s41598-025-02152-2