Med Phys. 2025 May 19. doi: 10.1002/mp.17895. Online ahead of print.

ABSTRACT

BACKGROUND: Microvascular invasion (MVI) is strongly associated with the prognosis of patients with hepatocellular carcinoma (HCC).

PURPOSE: To evaluate the value of Transformer models with Sonazoid contrast-enhanced ultrasound (CEUS) in the preoperative prediction of MVI.

METHODS: This retrospective study included 164 HCC patients. Deep learning features and radiomic features were extracted from arterial and Kupffer phase images, alongside the collection of clinicopathological parameters. Normality was assessed using the Shapiro-Wilk test. The Mann‒Whitney U-test and least absolute shrinkage and selection operator algorithm were applied to screen features. Transformer, radiomic, and clinical prediction models for MVI were constructed with logistic regression. Repeated random splits followed a 7:3 ratio, with model performance evaluated over 50 iterations. The area under the receiver operating characteristic curve (AUC, 95% confidence interval [CI]), sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), decision curve, and calibration curve were used to evaluate the performance of the models. The DeLong test was applied to compare performance between models. The Bonferroni method was used to control type I error rates arising from multiple comparisons. A two-sided p-value of < 0.05 was considered statistically significant.

RESULTS: In the training set, the diagnostic performance of the arterial-phase Transformer (AT) and Kupffer-phase Transformer (KT) models were better than that of the radiomic and clinical (Clin) models (p < 0.0001). In the validation set, both the AT and KT models outperformed the radiomic and Clin models in terms of diagnostic performance (p < 0.05). The AUC (95% CI) for the AT model was 0.821 (0.72-0.925) with an accuracy of 80.0%, and the KT model was 0.859 (0.766-0.977) with an accuracy of 70.0%. Logistic regression analysis indicated that tumor size (p = 0.016) and alpha-fetoprotein (AFP) (p = 0.046) were independent predictors of MVI.

CONCLUSIONS: Transformer models using Sonazoid CEUS have potential for effectively identifying MVI-positive patients preoperatively.

PMID:40384312 | DOI:10.1002/mp.17895

J Phys Chem Lett. 2025 May 18:5244-5251. doi: 10.1021/acs.jpclett.5c00592. Online ahead of print.

ABSTRACT

Machine learning (ML) approaches have become ubiquitous in the search for new materials in recent years. Bayesian optimization (BO) based on Gaussian processes (GPs) has become a widely recognized approach in material exploration. However, feature engineering has critical impacts on the efficiency of GP-based BO, because GPs cannot automatically generate descriptors. To address this limitation, this study applies deep kernel learning (DKL), which combines a neural network with a GP, to BO. The efficiency of the DKL model was comparable to or significantly better than that of a standard GP in a data set of 922 oxide data sets, covering band gaps, ionic dielectric constants, and effective masses of electrons, as well as in experimental data sets, the band gaps of 610 hybrid organic-inorganic perovskite alloys. When searching for the alloy with the highest Curie temperature among 4560 alloys, the standard GP outperformed the DKL model because a strongly correlated descriptor of the Curie temperature could be directly utilized. Additionally, DKL supports transfer learning, which further enhances its efficiency. Thus, we believe that BO based on DKL paves the way for exploring diverse material spaces more effectively than GPs.

PMID:40383929 | DOI:10.1021/acs.jpclett.5c00592

Sci Data. 2025 May 18;12(1):814. doi: 10.1038/s41597-025-05177-4.

ABSTRACT

Understanding human fertility requires dynamic and three-dimensional (3D) analysis of sperm movement, which extends beyond the capabilities of traditional datasets focused primarily on two-dimensional sperm motility or static morphological characteristics. To address this limitation, we introduce the 3D+t Multifocal Imaging Dataset of Human Sperm (3D-SpermVid), a repository comprising 121 multifocal video-microscopy hyperstacks of freely swimming sperm cells, incubated under non-capacitating conditions (NCC) and capacitating conditions (CC). This collection enables detailed observation and analysis of 3D sperm flagellar motility patterns over time, offering novel insights into the capacitation process and its implications for fertility. Data were captured using a multifocal imaging (MFI) system based on an optical microscope equipped with a piezoelectric device that adjusts focus at various heights, recording sperm movement in a volumetric space. By making this data publicly available, we aim to enable applications in deep learning and pattern recognition to uncover hidden flagellar motility patterns, fostering significant advancements in understanding 3D sperm morphology and dynamics, and developing new diagnostic tools for assessing male fertility, as well as assisting in the self-organizaton mechanisms driving spontaneous motility and navigation in 3D.

PMID:40383860 | DOI:10.1038/s41597-025-05177-4

Sci Rep. 2025 May 18;15(1):17263. doi: 10.1038/s41598-025-99858-0.

ABSTRACT

The widespread availability of miniaturized wearable fitness trackers has enabled the monitoring of various essential health parameters. Utilizing wearable technology for precise emotion recognition during human and computer interactions can facilitate authentic, emotionally aware contextual communication. In this paper, an emotion recognition system is proposed for the first time to conduct an experimental analysis of both discrete and dimensional models. An ensemble deep learning architecture is considered that consists of Long Short-Term Memory and Gated Recurrent Unit models to capture dynamic temporal dependencies within emotional data sequences effectively. The publicly available wearable devices EMOGNITION database is utilized to facilitate result reproducibility and comparison. The database includes physiological signals recorded using the Samsung Galaxy Watch, Empatica E4 wristband, and MUSE 2 Electroencephalogram (EEG) headband devices for a comprehensive understanding of emotions. A detailed comparison of all three dedicated wearable devices has been carried out to identify nine discrete emotions, exploring three different bio-signal combinations. The Samsung Galaxy and MUSE 2 devices achieve an average classification accuracy of 99.14% and 99.41%, respectively. The performance of the Samsung Galaxy device is examined for the 2D Valence-Arousal effective dimensional model. Results reveal average classification accuracy of 97.81% and 72.94% for Valence and Arousal dimensions, respectively. The acquired results demonstrate promising outcomes in emotion recognition when compared with the state-of-the-art methods.

PMID:40383809 | DOI:10.1038/s41598-025-99858-0

Sci Rep. 2025 May 18;15(1):17229. doi: 10.1038/s41598-025-02593-9.

ABSTRACT

Personalized recommendation systems are vital for enhancing user satisfaction and reducing information overload, especially in data-sparse environments like e-commerce platforms. This paper introduces a novel hybrid framework that combines Long Short-Term Memory (LSTM) with a modified Split-Convolution (SC) neural network (LSTM-SC) and an advanced sampling technique-Self-Inspected Adaptive SMOTE (SASMOTE). Unlike traditional SMOTE, SASMOTE adaptively selects "visible" nearest neighbors and incorporates a self-inspection strategy to filter out uncertain synthetic samples, ensuring high-quality data generation. Additionally, Quokka Swarm Optimization (QSO) and Hybrid Mutation-based White Shark Optimizer (HMWSO) are employed for optimizing sampling rates and hyperparameters, respectively. Experiments conducted on the goodbooks-10k and Amazon review datasets demonstrate significant improvements in RMSE, MAE, and R² metrics, proving the superiority of the proposed model over existing deep learning and collaborative filtering techniques. The framework is scalable, interpretable, and applicable across diverse domains, particularly in e-commerce and electronic publishing.

PMID:40383722 | DOI:10.1038/s41598-025-02593-9

Clin Nutr ESPEN. 2025 May 16:S2405-4577(25)00319-5. doi: 10.1016/j.clnesp.2025.05.022. Online ahead of print.

ABSTRACT

Enteral feeding needs secure access to the upper gastrointestinal tract, an evaluation of the gastric function to detect gastrointestinal intolerance, and a nutritional target to reach the patient's needs. Only in the last decades has progress been accomplished in techniques allowing an appropriate placement of the nasogastric tube, mainly reducing pulmonary complications. These techniques include point-of-care ultrasound (POCUS), electromagnetic sensors, real-time video-assisted placement, impedance sensors, and virtual reality. Again, POCUS is the most accessible tool available to evaluate gastric emptying, with antrum echo density measurement. Automatic measurements of gastric antrum content supported by deep learning algorithms and electric impedance provide gastric volume. Intragastric balloons can evaluate motility. Finally, advanced technologies have been tested to improve nutritional intake: Stimulation of the esophagus mucosa inducing contraction mimicking a contraction wave that may improve enteral nutrition efficacy, impedance sensors to detect gastric reflux and modulate the rate of feeding accordingly have been clinically evaluated. Use of electronic health records integrating nutritional needs, target, and administration is recommended.

PMID:40383254 | DOI:10.1016/j.clnesp.2025.05.022

J Cardiovasc Magn Reson. 2025 May 16:101913. doi: 10.1016/j.jocmr.2025.101913. Online ahead of print.

ABSTRACT

BACKGROUND: Image reconstruction from highly undersampled 4D flow MRI data can be very time consuming and may result in significant underestimation of velocities depending on regularization, thereby limiting the applicability of the method. The objective of the present work was to develop a generalizable self-supervised deep learning-based framework for fast and accurate reconstruction of highly undersampled 4D flow MRI and to demonstrate the utility of the framework for aortic and cerebrovascular applications.

METHODS: The proposed deep-learning-based framework, called FlowMRI-Net, employs physics-driven unrolled optimization using a complex-valued convolutional recurrent neural network and is trained in a self-supervised manner. The generalizability of the framework is evaluated using aortic and cerebrovascular 4D flow MRI acquisitions acquired on systems from two different vendors for various undersampling factors (R=8,16,24) and compared to compressed sensing (CS-LLR) reconstructions. Evaluation includes an ablation study and a qualitative and quantitative analysis of image and velocity magnitudes.

RESULTS: FlowMRI-Net outperforms CS-LLR for aortic 4D flow MRI reconstruction, resulting in significantly lower vectorial normalized root mean square error and mean directional errors for velocities in the thoracic aorta. Furthermore, the feasibility of FlowMRI-Net's generalizability is demonstrated for cerebrovascular 4D flow MRI reconstruction. Reconstruction times ranged from 3 to 7minutes on commodity CPU/GPU hardware.

CONCLUSION: FlowMRI-Net enables fast and accurate reconstruction of highly undersampled aortic and cerebrovascular 4D flow MRI, with possible applications to other vascular territories.

PMID:40383184 | DOI:10.1016/j.jocmr.2025.101913

Comput Biol Med. 2025 May 17;192(Pt B):110322. doi: 10.1016/j.compbiomed.2025.110322. Online ahead of print.

ABSTRACT

The application of AI for predicting critical heart failure endpoints using echocardiography is a promising avenue to improve patient care and treatment planning. However, fully supervised training of deep learning models in medical imaging requires a substantial amount of labelled data, posing significant challenges due to the need for skilled medical professionals to annotate image sequences. Our study addresses this limitation by exploring the potential of self-supervised learning, emphasising interpretability, robustness, and safety as crucial factors in cardiac imaging analysis. We leverage self-supervised learning on a large unlabelled dataset, facilitating the discovery of features applicable to a various downstream tasks. The backbone model not only generates informative features for training smaller models using simple techniques but also produces features that are interpretable by humans. The study employs a modified Self-supervised Transformer with Energy-based Graph Optimisation (STEGO) network on top of self-DIstillation with NO labels (DINO) as a backbone model, pre-trained on diverse medical and non-medical data. This approach facilitates the generation of self-segmented outputs, termed "parcels", which identify distinct anatomical sub-regions of the heart. Our findings highlight the robustness of these self-learned parcels across diverse patient profiles and phases of the cardiac cycle phases. Moreover, these parcels offer high interpretability and effectively encapsulate clinically relevant cardiac substructures. We conduct a comprehensive evaluation of the proposed self-supervised approach on publicly available datasets, demonstrating its adaptability to a wide range of requirements. Our results underscore the potential of self-supervised learning to address labelled data scarcity in medical imaging, offering a path to improve cardiac imaging analysis and enhance the efficiency and interpretability of diagnostic procedures, thus positively impacting patient care and clinical decision-making.

PMID:40383057 | DOI:10.1016/j.compbiomed.2025.110322

Epilepsy Behav. 2025 May 17;170:110470. doi: 10.1016/j.yebeh.2025.110470. Online ahead of print.

ABSTRACT

This study aimed to classify patients' focal (frontal, temporal, parietal, occipital), multifocal, and generalized epileptiform activities based on EEG findings using artificial intelligence models. The study included 575 patients followed in the Neurology Epilepsy Polyclinics of Adana City Training and Research Hospital between June 2021 and July 2024. Patient history, examination findings, seizure characteristics and EEG results were retrospectively reviewed to create a comprehensive database. Initially, machine learning architectures were applied to differentiate generalized and focal epilepsy. Subsequently, EEG findings were categorized into eight subgroups, and machine learning methods were utilized for classification. Three AI models-Multilayer Perceptron (MLP), Random Forest, and Support Vector Machine (SVM)-were employed. The dataset was further improved through data augmentation with SMOTE. The initial deep learning model achieved 89 % accuracy, recall, and F1 scores. Then, Optuna framework was incorporated into model to optimize hyperparameters, thus the accuracy reached 96 %. In comparison, the proposed ensemble model combining MLP, SVM and XGBoost achieved the highest accuracy of 98 %. The study demonstrates that data augmentation and ensemble AI models can provide robust decision support for physicians in classifying epilepsy types.

PMID:40382997 | DOI:10.1016/j.yebeh.2025.110470

Neural Netw. 2025 May 16;189:107573. doi: 10.1016/j.neunet.2025.107573. Online ahead of print.

ABSTRACT

In the field of healthcare, the acquisition and annotation of medical images present significant challenges, resulting in a scarcity of trainable samples. This data limitation hinders the performance of deep learning models, creating bottlenecks in clinical applications. To address this issue, we construct a framework (EScarcityS) aimed at enhancing the success rate of disease diagnosis in scarcity of trainable medical image scenarios. Firstly, considering that Transformer-based deep learning networks rely on a large amount of trainable data, this study takes into account the unique characteristics of pathological regions. By extracting the feature representations of all particles in medical images at different granularities, a multi-granularity Transformer network (MGVit) is designed. This network leverages additional prior knowledge to assist the Transformer network during training, thereby reducing the data requirement to some extent. Next, the importance maps of particles at different granularities, generated by MGVit, are fused to construct disease probability maps corresponding to the images. Based on these maps, a disease probability map-guided diffusion generation model is designed to generate more realistic and interpretable synthetic data. Subsequently, authentic and synthetical data are mixed and used to retrain MGVit, aiming to enhance the accuracy of medical image classification in scarcity of trainable medical image scenarios. Finally, we conducted detailed experiments on four real medical image datasets to validate the effectiveness of EScarcityS and its specific modules.

PMID:40382989 | DOI:10.1016/j.neunet.2025.107573

Water Res. 2025 May 13;283:123801. doi: 10.1016/j.watres.2025.123801. Online ahead of print.

ABSTRACT

Microalgae-driven nutrient recovery represents a promising technology for phosphorus removal from wastewater while simultaneously generating biomass that can be valorized to offset treatment costs. As full-scale processes come online, system parameters including biomass composition must be carefully monitored to optimize performance and prevent culture crashes. In this study, flow imaging microscopy (FIM) was leveraged to characterize microalgal community composition in near real-time at a full-scale municipal wastewater treatment plant (WWTP) in Wisconsin, USA, and population and morphotype dynamics were examined to identify relationships between water chemistry, biomass composition, and system performance. Two FIM technologies, FlowCam and ARTiMiS, were evaluated as monitoring tools. ARTiMiS provided a more accurate estimate of total system biomass, and estimates derived from particle area as a proxy for biovolume yielded better approximations than particle counts. Deep learning classification models trained on annotated image libraries demonstrated equivalent performance between FlowCam and ARTiMiS, and convolutional neural network (CNN) classifiers proved significantly more accurate when compared to feature table-based dense neural network (DNN) models. Across a two-year study period, Scenedesmus spp. appeared most important for phosphorus removal, and were negatively impacted by elevated temperatures and increase in nitrite/nitrate concentrations. Chlorella and Monoraphidium also played an important role in phosphorus removal. For both Scenedesmus and Chlorella, smaller morphological types were more often associated with better system performance, whereas larger morphotypes likely associated with stress response(s) correlated with poor phosphorus recovery rates. These results demonstrate the potential of FIM as a critical technology for high-resolution characterization of industrial microalgal processes.

PMID:40382876 | DOI:10.1016/j.watres.2025.123801

Comput Methods Programs Biomed. 2025 May 12;268:108812. doi: 10.1016/j.cmpb.2025.108812. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Diabetes is a chronic disease characterised by a high risk of developing diabetic nephropathy. The early identification of individuals at heightened risk of such complications or their exacerbation can be crucial to set a correct course of treatment. However, there are currently no widely accepted predictive tools for this task and, additionally, most of these models rely only on information at a single baseline visit. Considering this, we investigate the potential predictive role of patients' clinical history over multiple levels of renal disease severity while, at the same time, developing an effective predictive model.

METHODS: From the data collected in the DARWIN-Renal (DApagliflozin Real-World evIdeNce-Renal) study, a nationwide multicentre retrospective real-world study, we develop four different types of machine learning models, namely, logistic regression, random forest, Cox proportional hazards regression, and a deep learning model based on recurrent neural network to predict the crossing of 5 clinically relevant glomerular filtration rate thresholds for patients with type 2 diabetes.

RESULTS: The predictive performance of all models is satisfactory for all outcomes, even without the introduction of information referring to past visits, with AUROC and C-index between 0.69 and 0.98 and average precision well above the random model. The introduction of past information results into a clear improvement in performance for all the models, with percentage increases of up to 12% for both AUROC and C-index and 300% for average precision. The usefulness of past information is further corroborated by a feature importance analysis.

CONCLUSIONS: Incorporating data from the patients' clinical history into the predictive models greatly improves their performance, particularly for recurrent neural network where the full sequence of values for dynamic variables is provided compared to synthetic indicators of past history.

PMID:40382871 | DOI:10.1016/j.cmpb.2025.108812

Cell Rep. 2025 May 17;44(6):115715. doi: 10.1016/j.celrep.2025.115715. Online ahead of print.

ABSTRACT

Plants are one of the major reservoirs of medicinal compounds, serving as a cornerstone of both traditional and modern medicine. However, despite their importance, the complex biosynthetic pathways of many plant-derived compounds remain only partially understood, hindering their full potential in therapeutic applications. This review paper summarizes the advances in systems and synthetic biology utilized in the characterization and engineering of plant metabolic pathways. We discuss various strategies such as (1) co-expression analysis, (2) gene cluster identification, (3) metabolite profiling, (4) deep learning approaches, (5) genome-wide association studies, and (6) protein complex identification. Through case studies on several biosynthesis pathways, we highlight how these methods are applied to unravel complex pathways and enhance the production of important natural products. Finally, we discuss future directions in the context of metabolic engineering, including metabolon engineering, AI integration, and sustainable production strategies, underscoring the potential for cheaper and greener production of plant natural products.

PMID:40382775 | DOI:10.1016/j.celrep.2025.115715

Insights Imaging. 2025 May 18;16(1):108. doi: 10.1186/s13244-025-01988-6.

ABSTRACT

OBJECTIVES: Lymphovascular invasion significantly impacts the prognosis of urothelial carcinoma of the bladder. Traditional lymphovascular invasion detection methods are time-consuming and costly. This study aims to develop a deep learning-based model to preoperatively predict lymphovascular invasion status in urothelial carcinoma of bladder using CT images.

METHODS: Data and CT images of 577 patients across four medical centers were retrospectively collected. The largest tumor slices from the transverse, coronal, and sagittal planes were selected and used to train CNN models (InceptionV3, DenseNet121, ResNet18, ResNet34, ResNet50, and VGG11). Deep learning features were extracted and visualized using Grad-CAM. Principal Component Analysis reduced features to 64. Using the extracted features, Decision Tree, XGBoost, and LightGBM models were trained with 5-fold cross-validation and ensembled in a stacking model. Clinical risk factors were identified through logistic regression analyses and combined with DL scores to enhance lymphovascular invasion prediction accuracy.

RESULTS: The ResNet50-based model achieved an AUC of 0.818 in the validation set and 0.708 in the testing set. The combined model showed an AUC of 0.794 in the validation set and 0.767 in the testing set, demonstrating robust performance across diverse data.

CONCLUSION: We developed a robust radiomics model based on deep learning features from CT images to preoperatively predict lymphovascular invasion status in urothelial carcinoma of the bladder. This model offers a non-invasive, cost-effective tool to assist clinicians in personalized treatment planning.

CRITICAL RELEVANCE STATEMENT: We developed a robust radiomics model based on deep learning features from CT images to preoperatively predict lymphovascular invasion status in urothelial carcinoma of the bladder.

KEY POINTS: We developed a deep learning feature-based stacking model to predict lymphovascular invasion in urothelial carcinoma of the bladder patients using CT. Max cross sections from three dimensions of the CT image are used to train the CNN model. We made comparisons across six CNN networks, including ResNet50.

PMID:40382748 | DOI:10.1186/s13244-025-01988-6

ACS Sens. 2025 May 18. doi: 10.1021/acssensors.4c03397. Online ahead of print.

ABSTRACT

Surface-enhanced Raman spectroscopy (SERS) is a transformative tool for infectious disease diagnostics, offering rapid and sensitive species identification. However, background spectra in biological samples complicate analyte peak detection, increase the limit of detection, and hinder data augmentation. To address these challenges, we developed a deep learning framework utilizing dual neural networks to extract true virus SERS spectra and estimate concentration coefficients in water for 12 different respiratory viruses. The extracted spectra showed a high similarity to those obtained at the highest viral concentration, validating their accuracy. Using these spectra and the derived concentration coefficients, we augmented spectral data sets across varying virus concentrations in water. XGBoost models trained on these augmented data sets achieved overall classification and concentration prediction accuracy of 92.3% with a coefficient of determination (R2) > 0.95. Additionally, the extracted spectra and coefficients were used to augment data sets in saliva backgrounds. When tested against real virus-in-saliva spectra, the augmented spectra-trained XGBoost models achieved 91.9% accuracy in classification and concentration prediction with R2 > 0.9, demonstrating the robustness of the approach. By delivering clean and uncontaminated spectra, this methodology can significantly improve species identification, differentiation, and quantification and advance SERS-based detection and diagnostics.

PMID:40382719 | DOI:10.1021/acssensors.4c03397

Arch Iran Med. 2025 Apr 1;28(4):198-206. doi: 10.34172/aim.32070. Epub 2025 Apr 1.

ABSTRACT

BACKGROUND: To investigate whether the bone age (BA) of Iranian children could be accurately assessed via an artificial intelligence (AI) system. Accurate assessment of skeletal maturity is crucial for diagnosing and treating various musculoskeletal disorders, and is traditionally achieved through manual comparison with the Greulich-Pyle atlas. This process, however, is subjective and time-consuming. Recent advances in deep learning offer more efficient and consistent BA evaluations.

METHODS: From left-hand radiographs of children aged 1-18 years who presented to a tertiary research hospital, 555 radiographs (220 boys and 335 girls) were collected. The reference BA was determined via the Greulich and Pyle (GP) method by two radiologists in consensus. The BA was then estimated to use a deep learning model specifically developed for this population. Model performance was evaluated using multiple metrics: Mean square error (MSE), mean absolute error (MAE), intra-class correlation coefficient (ICC), and 95% limits of agreement (LoA). Gender-specific results were analyzed separately.

RESULTS: The model demonstrated acceptable accuracy. For boys, MSE was 0.55 years, MAE was 0.59 years, ICC was 0.74, and the 95% LoA ranged from -0.8 to 1.2 years. For girls, MSE was 0.59 years, MAE was 0.61 years, ICC was 0.82, and the 95% LoA ranged from -0.6 to 1.0 years. These results indicate stronger predictive accuracy for girls compared to boys.

CONCLUSION: Our findings demonstrate that the proposed deep learning model achieves reasonable accuracy in BA assessment, with stronger performance in girls compared to boys. However, the relatively wide 95% LoA, particularly for boys, and prediction errors at the extremes of the age range highlight the need for further refinement and validation. While the model shows potential as a supplementary tool for clinicians, future studies should focus on improving prediction accuracy, reducing variability, and validating the model on larger, more diverse datasets before considering widespread clinical implementation. Additionally, addressing edge cases and specific conditions that a human reviewer may detect but the model might overlook, will be essential for enhancing its clinical reliability.

PMID:40382691 | DOI:10.34172/aim.32070

Plant Methods. 2025 May 17;21(1):63. doi: 10.1186/s13007-025-01381-w.

ABSTRACT

Pest infestations in rice crops greatly affect yield and quality, making early detection essential. As most rice pests affect leaves and rhizomes, visual inspection of rice for pests is becoming increasingly important. In precision agriculture, fast and accurate automatic pest identification is essential. To tackle this issue, multiple models utilizing computer vision and deep learning have been applied. Owing to its high efficiency, deep learning is now the favored approach for detecting plant pests. In this regard, the paper introduces an effective rice pest detection framework utilizing the Transformer architecture, designed to capture long-range features. The paper enhances the original model by adding the self-developed RepPConv-block to reduce the problem of information redundancy in feature extraction in the model backbone and to a certain extent reduce the model parameters. The original model's CCFM structure is enhanced by integrating the Gold-YOLO neck, improving its ability to fuse multi-scale features. Additionally, the MPDIoU-based loss function enhances the model's detection performance. Using the self-constructed high-quality rice pest dataset, the model achieves higher identification accuracy while reducing the number of parameters. The proposed RP18-DETR and RP34-DETR models reduce parameters by 16.5% and 25.8%, respectively, compared to the original RT18-DETR and RT34-DETR models. With a threshold of 0.5, the average accuracy calculated is 1.2% higher for RP18-DETR than for RT18-DETR.

PMID:40382633 | DOI:10.1186/s13007-025-01381-w

BMC Cardiovasc Disord. 2025 May 17;25(1):371. doi: 10.1186/s12872-025-04739-z.

ABSTRACT

Globally, the high number of deaths are happening due to Myocardial infarction (MI). MI is considered as a life-threatening disease, which leads to an increase number of deaths or damage to the heart, and hence, prompt detection of MI is critical to decrease the mortality rate. Though, numerous works have addressed MI identification, an increased number suffer from over fitting and high computational burden in real-time scenarios. The proposed system introduces a novel MI detection technique using a Deep Residual Network (DRN), where the solution is optimized by the proposed Social Ski-Spider (SSS) Optimization algorithm is the novel combination of both Social Ski-driver (SSD) Optimization and the Spider Monkey Optimization (SMO). This model highly prevents the overfitting and computational burden, which increases the MI detection accuracy. Here, the proposed SSS-DRN performs detection by filtering the electrocardiography (ECG) signals. Later, the signal feature, transform feature, medical feature and statistical feature are extracted by the feature extraction phase followed by data augmentation that consists of permutation, random generation and re-sampling processes and finally, detection is accomplished by the SSS-DRN. Moreover, the developed SSS-DRN is researched for its efficiency considering metrics like accuracy, sensitivity, and specificity and observed 0.916, 0.921, and 0.926. Here, when considering the accuracy metrics, the performance gain observed by the devised model is 13.96%, 12.61%, 10.37%, 7.95%, 5%, 2.21%, and 2% higher than the traditional schemes. This indicates the devised model has high detection accuracy, which could be embedded in real-time clinical settings like hospital ECG machines, wearable ECG monitors, and mobile health applications. This improves the clinical decision-making process with increased patient outcomes.

PMID:40382575 | DOI:10.1186/s12872-025-04739-z

NPJ Digit Med. 2025 May 17;8(1):291. doi: 10.1038/s41746-025-01641-y.

ABSTRACT

Deep learning techniques have significantly enhanced the convenience and precision of ultrasound image diagnosis, particularly in the crucial step of lesion segmentation. However, recent studies reveal that both train-from-scratch models and pre-trained models often exhibit performance disparities across sex and age attributes, leading to biased diagnoses for different subgroups. In this paper, we propose APPLE, a novel approach designed to mitigate unfairness without altering the parameters of the base model. APPLE achieves this by learning fair perturbations in the latent space through a generative adversarial network. Extensive experiments on both a publicly available dataset and an in-house ultrasound image dataset demonstrate that our method improves segmentation and diagnostic fairness across all sensitive attributes and various backbone architectures compared to the base models. Through this study, we aim to highlight the critical importance of fairness in medical segmentation and contribute to the development of a more equitable healthcare system.

PMID:40382499 | DOI:10.1038/s41746-025-01641-y

Eur Radiol. 2025 May 17. doi: 10.1007/s00330-025-11666-2. Online ahead of print.

ABSTRACT

OBJECTIVES: This study aims to develop a deep learning algorithm for differentiating aneurysmal subarachnoid hemorrhage (aSAH) from non-aneurysmal subarachnoid hemorrhage (naSAH) using non-contrast computed tomography (NCCT) scans.

METHODS: This retrospective study included 618 patients diagnosed with SAH. The dataset was divided into a training and internal validation cohort (533 cases: aSAH = 305, naSAH = 228) and an external test cohort (85 cases: aSAH = 55, naSAH = 30). Hemorrhage regions were automatically segmented using a U-Net + + architecture. A ResNet-based deep learning model was trained to classify the etiology of SAH.

RESULTS: The model achieved robust performance in distinguishing aSAH from naSAH. In the internal validation cohort, it yielded an average sensitivity of 0.898, specificity of 0.877, accuracy of 0.889, Matthews correlation coefficient (MCC) of 0.777, and an area under the curve (AUC) of 0.948 (95% CI: 0.929-0.967). In the external test cohort, the model demonstrated an average sensitivity of 0.891, specificity of 0.880, accuracy of 0.887, MCC of 0.761, and AUC of 0.914 (95% CI: 0.889-0.940), outperforming junior radiologists (average accuracy: 0.836; MCC: 0.660).

CONCLUSION: The study presents a deep learning architecture capable of accurately identifying SAH etiology from NCCT scans. The model's high diagnostic performance highlights its potential to support rapid and precise clinical decision-making in emergency settings.

KEY POINTS: Question Differentiating aneurysmal from naSAH is crucial for timely treatment, yet existing imaging modalities are not universally accessible or convenient for rapid diagnosis. Findings A ResNet-variant-based deep learning model utilizing non-contrast CT scans demonstrated high accuracy in classifying SAH etiology and enhanced junior radiologists' diagnostic performance. Clinical relevance AI-driven analysis of non-contrast CT scans provides a fast, cost-effective, and non-invasive solution for preoperative SAH diagnosis. This approach facilitates early identification of patients needing aneurysm surgery while minimizing unnecessary angiography in non-aneurysmal cases, enhancing clinical workflow efficiency.

PMID:40382487 | DOI:10.1007/s00330-025-11666-2