J Med Imaging (Bellingham). 2025 Jan;12(1):014003. doi: 10.1117/1.JMI.12.1.014003. Epub 2025 Feb 3.

ABSTRACT

PURPOSE: The restoration of images affected by blur and noise has been widely studied and has broad potential for applications including in medical imaging modalities such as computed tomography. Recently, deep learning approaches have demonstrated the potential to enhance image quality beyond classic limits; however, most deep learning models attempt a blind restoration problem and base their restoration on image inputs alone without direct knowledge of the image noise and blur properties. We present a method that leverages both degraded image inputs and a characterization of the system's blur and noise to combine modeling and deep learning approaches.

APPROACH: Different methods to integrate these auxiliary inputs are presented, namely, an input-variant and a weight-variant approach wherein the auxiliary inputs are incorporated as a parameter vector before and after the convolutional block, respectively, allowing easy integration into any convolutional neural network architecture.

RESULTS: The proposed model shows superior performance compared with baseline models lacking auxiliary inputs. Evaluations are based on the average peak signal-to-noise ratio and structural similarity index measure, selected examples of top and bottom 10% performance for varying approaches, and an input space analysis to assess the effect of different noise and blur on performance. In addition, the proposed model exhibits a degree of robustness when the blur and noise parameters deviate from their true values.

CONCLUSION: Results demonstrate the efficacy of providing a deep learning model with auxiliary inputs, representing system blur and noise characteristics, to enhance the performance of the model in image restoration tasks.

PMID:39906485 | PMC:PMC11788843 | DOI:10.1117/1.JMI.12.1.014003

Front Bioeng Biotechnol. 2025 Jan 21;13:1506508. doi: 10.3389/fbioe.2025.1506508. eCollection 2025.

ABSTRACT

Protein function prediction is crucial in several key areas such as bioinformatics and drug design. With the rapid progress of deep learning technology, applying protein language models has become a research focus. These models utilize the increasing amount of large-scale protein sequence data to deeply mine its intrinsic semantic information, which can effectively improve the accuracy of protein function prediction. This review comprehensively combines the current status of applying the latest protein language models in protein function prediction. It provides an exhaustive performance comparison with traditional prediction methods. Through the in-depth analysis of experimental results, the significant advantages of protein language models in enhancing the accuracy and depth of protein function prediction tasks are fully demonstrated.

PMID:39906415 | PMC:PMC11790633 | DOI:10.3389/fbioe.2025.1506508

J Imaging Inform Med. 2025 Feb 4. doi: 10.1007/s10278-025-01428-3. Online ahead of print.

ABSTRACT

Vaginitis is a common condition in women that is described medically as irritation and/or inflammation of the vagina; it poses a significant health risk for women, necessitating precise diagnostic methods. Presently, conventional techniques for examining vaginal discharge involve the use of wet mounts and gram staining to identify vaginal diseases. In this research, we utilized fluorescent staining, which enables distinct visualization of cellular and pathogenic components, each exhibiting unique color characteristics when exposed to the same light source. We established a large, challenging multiple fluorescence leucorrhea dataset benchmark comprising 8 categories with a total of 343 K high-quality labels. We also presented a robust lightweight deep-learning network, LRNet. It includes a lightweight feature extraction network that employs Ghost modules, a feature pyramid network that incorporates deformable convolution in the neck, and a single detection head. The evaluation results indicate that this detection network surpasses conventional networks and can cut down the model parameters by up to 91.4% and floating-point operations (FLOPs) by 74%. The deep-optimal leucorrhea detection capability of LRNet significantly enhances its ability to detect various crucial indicators related to vaginal health.

PMID:39904942 | DOI:10.1007/s10278-025-01428-3

J Imaging Inform Med. 2025 Feb 4. doi: 10.1007/s10278-024-01376-4. Online ahead of print.

ABSTRACT

The study aimed to predict expression of pituitary transcription factor 1 (PIT1) in pituitary adenomas using habitat, intra-tumoral and peri-tumoral radiomics models. A total of 129 patients with pituitary adenoma (training set, n = 103; test set, n = 26) were retrospectively enrolled. A total of 12, 18, 14, 13, and 14 radiomics features were selected from the ROIintra, ROIintra+peri (ROIintra+2mm, ROIintra+4mm, ROIintra+6mm), and ROIhabitat, respectively. Then, three machine learning algorithms were employed to develop radiomic models, including logistic regression (LR), support vector machines (SVM), and multilayer perceptron (MLP). The performances of the intra-tumoral, combined intra-tumoral and peri-tumoral, and habitat models were evaluated. The peritumoral region (ROI2mm, ROI4mm, ROI6mm) of the combined model with the highest performance was individually selected for further peritumoral analysis. Moreover, a deep learning radiomics nomogram (DLRN) was constructed incorporating clinical characteristics and the peri-tumoral and habitat models for individual prediction. The combined modelintra+2mm based on ROIintra+2mm achieved a better performance (AUC, 0.800) than that of the intra-tumoral model alone (AUC, 0.731). And the habitat model showed a higher performance (AUC, 0.806) than that of the intra-tumoral model. In addition, the performance of the peri-tumoral model based on ROI2mm was 0.694 in the testing set. Furthermore, the DLRN achieved the highest performance of 0.900 in the test set. The DLRN showed the best performance for PIT1 expression in pituitary adenomas, followed by the habitat, combined modelintra+2mm, intra-tumoral model, and peri-tumoral model based on ROI2mm, respectively. These different models are helpful for the model choice in clinical work.

PMID:39904941 | DOI:10.1007/s10278-024-01376-4

J Imaging Inform Med. 2025 Feb 4. doi: 10.1007/s10278-025-01419-4. Online ahead of print.

ABSTRACT

Diabetic retinopathy is a major complication of diabetes, with its prevalence nearly doubling to approximately 10.5% by 2021. Exudates, the characteristic lesions of diabetic retinopathy, are crucial for assessing disease progression and severity. The location and distribution of these exudates can affect various regions of the retina, necessitating a detailed regional analysis of lesions. To address this need, this study aimed to evaluate the performance of exudate detection in fundus images across various regions, including perivascular and extravascular areas, perifoveal and extrafoveal regions, and in quadrants defined relative to the fovea. We employed U-net and U-net3 + deep learning models for validation, evaluating their performance using accuracy, sensitivity, specificity, and Dice score. Overall, the U-net3 + model outperformed the U-net model. Therefore, the performance evaluation was based on the results from the U-net3 + model. Comparing the detection performance across perivascular versus extravascular and perifoveal versus extrafoveal regions, the U-net3 + model achieved highest Dice score in the extravascular (87.96% [± 5.80]) and perifoveal areas (88.03% [± 5.86]). Additionally, superior sensitivity and Dice scores were observed in the top-left and top-right quadrants. Future research is anticipated to show that deep learning-based automatic exudate detection will enhance diagnostic accuracy and efficiency, leading to better treatment and prognosis in patients with diabetic retinopathy.

PMID:39904940 | DOI:10.1007/s10278-025-01419-4

Neurocrit Care. 2025 Feb 4. doi: 10.1007/s12028-025-02217-0. Online ahead of print.

ABSTRACT

BACKGROUND: Despite advancements in the neuroscience of consciousness, no new medications for disorders of consciousness (DOC) have been discovered in more than a decade. Repurposing existing US Food and Drug Administration (FDA)-approved drugs for DOC is crucial for improving clinical management and patient outcomes.

METHODS: To identify potential new treatments among existing FDA-approved drugs, we used a deep learning-based drug screening model to predict the efficacy of drugs as awakening agents based on their three-dimensional molecular structure. A retrospective cohort study from March 2012 to October 2024 tested the model's predictions, focusing on changes in Glasgow Coma Scale (GCS) scores in 4047 patients in a coma from traumatic, vascular, or anoxic brain injury.

RESULTS: Our deep learning drug screens identified saxagliptin, a dipeptidyl peptidase-4 inhibitor, as a promising awakening drug for both acute and prolonged DOC. The retrospective clinical analysis showed that saxagliptin was associated with the highest recovery rate from acute coma among diabetes medications. After matching patients by age, sex, initial GCS score, coma etiology, and glycemic status, brain-injured patients with diabetes on incretin-based therapies, including dipeptidyl peptidase-4 inhibitors and glucagon-like peptide-1 analogues, recovered from coma at significantly higher rates compared to both brain-injured patients with diabetes on non-incretin-based diabetes medications (95% confidence interval of 1.8-14.1% higher recovery rate, P = 0.0331) and brain-injured patients without diabetes (95% confidence interval of 2-21% higher recovery rate, P = 0.0272). Post matching, brain-injured patients with diabetes on incretin-based therapies also recovered at a significantly higher rate than patients treated with amantadine (95% confidence interval for the difference 2.4-25.1.0%, P = 0.0364). A review of preclinical studies identified several pathways through which saxagliptin and other incretin-based medications may aid awakening from both acute and chronic DOC: restoring monoaminergic and GABAergic neurotransmission, reducing brain inflammation and oxidative damage, clearing hyperphosphorylated tau and amyloid-β, normalizing thalamocortical glucose metabolism, increasing neural plasticity, and mitigating excitotoxic brain damage.

CONCLUSIONS: Our findings suggest incretin-based medications in general, and saxagliptin in particular, as potential novel therapeutic agents for DOC. Further prospective clinical trials are needed to confirm their efficacy and safety in DOC.

PMID:39904872 | DOI:10.1007/s12028-025-02217-0

Genes Cells. 2025 Mar;30(2):e70000. doi: 10.1111/gtc.70000.

ABSTRACT

Cis-regulatory elements (cREs) play a crucial role in regulating gene expression and determining cell differentiation and state transitions. To capture the heterogeneous transitions of cell states associated with these processes, detecting cRE activity at the single-cell level is essential. However, current analytical methods can only capture the average behavior of cREs in cell populations, thereby obscuring cell-specific variations. To address this limitation, we proposed an attention-based deep neural network framework that integrates DNA sequences, genomic distances, and single-cell multi-omics data to detect cREs and their activities in individual cells. Our model shows higher accuracy in identifying cREs within single-cell multi-omics data from healthy human peripheral blood mononuclear cells than other existing methods. Furthermore, it clusters cells more precisely based on predicted cRE activities, enabling a finer differentiation of cell states. When applied to publicly available single-cell data from patients with glioma, the model successfully identified tumor-specific SOX2 activity. Additionally, it revealed the heterogeneous activation of the ZEB1 transcription factor, a regulator of epithelial-to-mesenchymal transition-related genes, which conventional methods struggle to detect. Overall, our model is a powerful tool for detecting cRE regulation at the single-cell level, which may contribute to revealing drug resistance mechanisms in cell sub-populations.

PMID:39904740 | DOI:10.1111/gtc.70000

Med Phys. 2025 Feb 4. doi: 10.1002/mp.17668. Online ahead of print.

ABSTRACT

BACKGROUND: Although deep learning (DL) methods for reconstructing 3D magnetic resonance (MR) volumes from 2D MR images yield promising results, they require large amounts of training data to perform effectively. To overcome this challenge, fine-tuning-a transfer learning technique particularly effective for small datasets-presents a robust solution for developing personalized DL models.

PURPOSE: A 2D to 3D conditional generative adversarial network (GAN) model with a patient- and fraction-specific fine-tuning workflow was developed to reconstruct synthetic 3D MR volumes using orthogonal 2D MR images for online dose adaptation.

METHODS: A total of 2473 3D MR volumes were collected from 43 patients. The training and test datasets were separated into 34 and 9 patients, respectively. All patients underwent MR-guided adaptive radiotherapy using the same imaging protocol. The population data contained 2047 3D MR volumes from the training dataset. Population data were used to train the population-based GAN model. For each fraction of the remaining patients, the population model was fine-tuned with the 3D MR volumes acquired before beam irradiation of the fraction, named the fine-tuned model. The performance of the fine-tuned model was tested using the 3D MR volume acquired immediately after the beam delivery of the fraction. The model's input was a pair of axial and sagittal MR images at the isocenter level, and the output was a 3D MR volume. Model performance was evaluated using the structural similarity index measure (SSIM), peak signal-to-noise ratio (PSNR), root mean square error (RMSE), and mean absolute error (MAE). Moreover, the prostate, bladder, and rectum in the predicted MR images were manually segmented. To assess geometric accuracy, the 2D Dice Similarity Coefficient (DSC) and 2D Hausdorff Distance (HD) were calculated.

RESULTS: A total of 84 3D MR volumes were included in the performance testing. The mean ± standard deviation (SD) of SSIM, PSNR, RMSE, and MAE were 0.64 ± 0.10, 93.9 ± 1.5 dB, 0.050 ± 0.009, and 0.036 ± 0.007 for the population model and 0.72 ± 0.09, 96.2 ± 1.8 dB, 0.041 ± 0.007, and 0.028 ± 0.006 for the fine-tuned model, respectively. The image quality of the fine-tuned model was significantly better than that of the population model (p < 0.05). The mean ± SD of DSC and HD of the population model were 0.79 ± 0.08 and 1.70 ± 2.35 mm for prostate, 0.81 ± 0.10 and 2.75 ± 1.53 mm for bladder, and 0.72 ± 0.08 and 1.93 ± 0.59 mm for rectum. Contrarily, the mean ± SD of DSC and HD of the fine-tuned model were 0.83 ± 0.06 and 1.29 ± 0.77 mm for prostate, 0.85 ± 0.07 and 2.16 ± 1.09 mm for bladder, and 0.77 ± 0.08 and 1.57 ± 0.52 mm for rectum. The geometric accuracy of the fine-tuned model was significantly improved than that of the population model (p < 0.05).

CONCLUSION: By employing a patient- and fraction-specific fine-tuning approach, the GAN model demonstrated promising accuracy despite limited data availability.

PMID:39904621 | DOI:10.1002/mp.17668

Med Phys. 2025 Feb 4. doi: 10.1002/mp.17663. Online ahead of print.

ABSTRACT

BACKGROUND: Kidney ultrasound (US) image segmentation is one of the key steps in computer-aided diagnosis and treatment planning of kidney diseases. Recently, deep learning (DL) technology has demonstrated promising prospects in automatic kidney US segmentation. However, due to the poor quality, particularly the weak boundaries in kidney US imaging, obtaining accurate annotations for DL-based segmentation methods remain a challenging and time-consuming task. This issue can hinder the application of data-hungry deep learning methods.

PURPOSE: In this paper, we explore a novel cross-modal data augmentation method aimed at enhancing the performance of DL-based segmentation networks on the limited labeled kidney US dataset.

METHODS: In particular, we adopt a novel method based on contrastive unpaired translation network (CUT) to obtain simulated labeled kidney US images at a low cost from labeled abdomen computed tomography (CT) data and unlabeled kidney US images. To effectively improve the segmentation network performance, we propose an instance-weighting training strategy that simultaneously captures useful information from both the simulated and real labeled kidney US images. We trained our generative networks on a dataset comprising 4418 labeled CT slices and 4594 unlabeled US images. For segmentation network, we used a dataset consisting of 4594 simulated and 100 real kidney US images for training, 20 images for validation, and 169 real images for testing. We compared the performance of our method to several state-of-the-art approaches using the Wilcoxon signed-rank test, and applied the Bonferroni method for multiple comparison correction.

RESULTS: The experimental results show that we can synthesize accurate labeled kidney US images with a Fréchet inception distance of 52.52. Moreover, the proposed method achieves a segmentation accuracy of 0.9360 ± 0.0398 for U-Net on normal kidney US images, and 0.7719 ± 0.2449 on the abnormal dataset, as measured by the dice similarity coefficient. When compared to other training strategies, the proposed method demonstrated statistically significant superiority, with all p-values being less than 0.01.

CONCLUSIONS: The proposed method can effectively improve the accuracy and generalization ability of kidney US image segmentation models with limited annotated training data.

PMID:39904615 | DOI:10.1002/mp.17663

Brain Res. 2025 Feb 2:149484. doi: 10.1016/j.brainres.2025.149484. Online ahead of print.

ABSTRACT

Accurate recognition and classification of motor imagery electroencephalogram (MI-EEG) signals are crucial for the successful implementation of brain-computer interfaces (BCI). However, inherent characteristics in original MI-EEG signals, such as nonlinearity, low signal-to-noise ratios, and large individual variations, present significant challenges for MI-EEG classification using traditional machine learning methods. To address these challenges, we propose an automatic feature extraction method rooted in deep learning for MI-EEG classification. First, original MI-EEG signals undergo noise reduction through discrete wavelet transform and common average reference. To reflect the regularity and specificity of brain neural activities, a convolutional neural network (CNN) is used to extract the time-domain features of MI-EEG. We also extracted spatial features to reflect the activity relationships and connection states of the brain in different regions. This process yields time series containing spatial information, focusing on enhancing crucial feature sequences through talking-heads attention. Finally, more abstract spatial-temporal features are extracted using a temporal convolutional network (TCN), and classification is done through a fully connected layer. Validation experiments based on the BCI Competition IV-2a dataset show that the enhanced EEG model achieves an impressive average classification accuracy of 85.53% for each subject. Compared with CNN, EEGNet, CNN-LSTM and EEG-TCNet, the classification accuracy of this model is improved by 11.24%, 6.90%, 11.18% and 6.13%, respectively. Our work underscores the potential of the proposed model to enhance intention recognition in MI-EEG significantly.

PMID:39904453 | DOI:10.1016/j.brainres.2025.149484

Tuberc Respir Dis (Seoul). 2025 Feb 4. doi: 10.4046/trd.2024.0166. Online ahead of print.

ABSTRACT

BACKGROUND: Computed tomography (CT) scans are used to assess emphysema, a significant phenotype of chronic obstructive pulmonary disease (COPD), but variability in CT protocols and devices across the hospitals may affect accuracy. This study aims to perform kernel conversion among different CT settings and to evaluate differences in the correlation between emphysema index before and after kernel conversion, as well as clinical measures in COPD patients.

METHODS: The data were extracted from the Korea COPD Subgroup Study database, involving 484 COPD patients with CT scan images. These were processed with kernel conversion. Emphysema extent was quantified as the percentage of low-attenuation areas (%LAA-950) by deep learning-based program. The correlation between %LAA-950 and clinical parameters, such as lung function tests, the modified Medical Research Council (mMRC), six-minute walking distance (6MWD), COPD assessment test (CAT), and the St. George's Respiratory Questionnaire for COPD (SGRQ-c), were analyzed. These values were then compared across different CT settings.

RESULTS: A total of 484 participants were included. Compared to before, kernel conversion reduced the variance in %LAA-950 values (before vs. after: 12.6±11.0 vs. 8.8±11.9). After kernel conversion, %LAA-950 showed moderate correlations with forced expiratory volume in one second (r = -0.41), residual volume/total lung capacity (r = 0.42), mMRC (r = 0.25), CAT score (r = 0.12), SGRQ-c (r = 0.21), and 6MWD (r = 0.15), all of which improved compared to the unconverted dataset (all, P<0.01).

CONCLUSION: CT images processed with kernel conversion improve the correlation between emphysema extent and clinical parameters in COPD.

PMID:39904364 | DOI:10.4046/trd.2024.0166

Nurse Educ Today. 2025 Jan 31;147:106600. doi: 10.1016/j.nedt.2025.106600. Online ahead of print.

ABSTRACT

OBJECTIVE: This umbrella review of reviews aims to synthesise current evidence on AI's utilisation in developing education within health sciences disciplines.

DESIGN: An umbrella review of reviews, review of reviews, based on Joanna Briggs Institute guidelines.

DATA SELECTION: CINAHL, ERIC(ProQuest), PubMed, Scopus, and Medic were systematically searched in December 2023 with no time limit. The inclusion and exclusion criteria were defined according to the PCC framework: Participants(P), Concept(C), and Context (C). Two independent researchers screened 6304 publications, and 201 reviews were selected in the full-text phase.

DATA EXTRACTION: All the reviews that met inclusion criteria were included in the analysis. The reference lists of included reviews were also searched. Included reviews were quality appraised. The results were analysed with narrative synthesis.

RESULTS OF DATA SYNTHESIS: Seven reviews published between 2019 and 2023 were selected for analysis. Five key domains were identified: robotics, machine learning and deep learning, big data, immersive technologies, and natural language processing. Robotics enhances practical medical, dental and nursing education training. Machine learning personalises learning experiences and improves diagnostic skills. Immersive technologies provide interactive simulations for practical training.

CONCLUSION: This umbrella review of reviews highlights the potential of AI in health sciences education and the need for continued investment in AI technologies and ethical frameworks to ensure effective and equitable integration into educational practices.

PMID:39904286 | DOI:10.1016/j.nedt.2025.106600

Evol Comput. 2025 Feb 4:1-27. doi: 10.1162/evco_a_00367. Online ahead of print.

ABSTRACT

In many recent works,the potential of Exploratory Landscape Analysis (ELA) features to numerically characterize single-objective continuous optimization problems has been demonstrated. These numerical features provide the input for all kinds of machine learning tasks in the domain of continuous optimization problems, ranging, i.a., from High-level Property Prediction to Automated Algorithm Selection and Automated Algorithm Configuration. Without ELA features, analyzing and understanding the characteristics of single-objective continuous optimization problems is - to the best of our knowledge - very limited. Yet, despite their usefulness, as demonstrated in several past works, ELA features suffer from several drawbacks. These include, in particular, (1.) a strong correlation between multiple features, as well as (2.) its very limited applicability to multiobjective continuous optimization problems. As a remedy, recent works proposed deep learning-based approaches as alternatives to ELA. In these works, among others point-cloud transformers were used to characterize an optimization problem's fitness landscape. However, these approaches require a large amount of labeled training data. Within this work, we propose a hybrid approach, Deep-ELA, which combines (the benefits of) deep learning and ELA features. We pre-trained four transformers on millions of randomly generated optimization problems to learn deep representations of the landscapes of continuous single- and multi-objective optimization problems. Our proposed framework can either be used out-of-the-box for analyzing single- and multiobjective continuous optimization problems, or subsequently fine-tuned to various tasks focusing on algorithm behavior and problem understanding.

PMID:39903851 | DOI:10.1162/evco_a_00367

ACS Nano. 2025 Feb 4. doi: 10.1021/acsnano.4c15364. Online ahead of print.

ABSTRACT

Understanding molecular dynamics at the nanoscale remains challenging due to limitations in the temporal resolution of current imaging techniques. Deep learning integrated with Single-Molecule Localization Microscopy (SMLM) offers opportunities to probe these dynamics. Here, we leverage this integration to reveal entangled polymer dynamics at a fast time scale, which is relatively poorly understood at the single-molecule level. We used Lambda DNA as a model system and modeled their entanglement using the self-avoiding wormlike chain model, generated simulated localizations along the contours, and trained the deep learning algorithm on these simulated images to predict chain contours from sparse localization data. We found that the localizations are heterogeneously distributed along the contours. Our assessments indicated that chain entanglement creates local diffusion barriers for switching buffer molecules, affecting the photoswitching kinetics of fluorescent dyes conjugated to the DNA molecules at discrete DNA segments. Tracking these segments demonstrated stochastic and subdiffusive migration of the entanglement loci. Our approach provides direct visualization of nanoscale polymer dynamics and local molecular environments previously inaccessible to conventional imaging techniques. In addition, our results suggest that the switching kinetics of the fluorophores in SMLM can be used to characterize nanoscopic local environments.

PMID:39903818 | DOI:10.1021/acsnano.4c15364

PLoS One. 2025 Feb 4;20(2):e0318550. doi: 10.1371/journal.pone.0318550. eCollection 2025.

ABSTRACT

Detecting cracks in asphalt concrete slabs is challenging due to environmental factors like lighting changes, surface reflections, and weather conditions, which affect image quality and crack detection accuracy. This study introduces a novel deep learning-based anomaly model for effective crack detection. A large dataset of panel images was collected and processed using denoising, standardization, and data augmentation techniques, with crack areas labeled via LabelImg software. The core model is an improved Xception network, enhanced with an adaptive activation function, dynamic attention mechanism, and multi-level residual connections. These innovations optimize feature extraction, enhance feature weighting, and improve information transmission, significantly boosting accuracy and robustness. The improved model achieves a 97.6% accuracy and a Matthews correlation coefficient of 0.98, remaining stable under varying lighting conditions. This method not only provides a fresh approach to crack detection but also greatly enhances detection efficiency.

PMID:39903732 | DOI:10.1371/journal.pone.0318550

PLoS One. 2025 Feb 4;20(2):e0317136. doi: 10.1371/journal.pone.0317136. eCollection 2025.

ABSTRACT

Adenomyosis and leiomyomas are common benign uterine disorders characterized by abnormal cellular proliferation. The BCL6 protein, a transcriptional repressor implicated in cell proliferation and oncogenesis, has been linked to the pathogenesis of endometriosis. This study investigates BCL6 expression in adenomyosis, leiomyomas, and normal myometrium using immunohistochemistry and deep learning neural networks. We analyzed paraffin blocks from total hysterectomies performed between 2009 and 2017, confirming diagnoses through pathological review. Immunohistochemistry was conducted using an automated system, and BCL6 expression was quantified using Fiji-ImageJ software. A supervised deep learning neural network was employed to classify samples based on DAB staining. Our results show that BCL6 expression is significantly higher in leiomyomas compared to adenomyosis and normal myometrium. No significant difference in BCL6 expression was observed between adenomyosis and controls. The deep learning neural network accurately classified samples with a high degree of precision, supporting the immunohistochemical findings. These findings suggest that BCL6 plays a role in the pathogenesis of leiomyomas, potentially contributing to abnormal smooth muscle cell proliferation. The study highlights the utility of automated immunohistochemistry and deep learning techniques in quantifying protein expression and classifying uterine pathologies. Future studies should investigate the expression of BCL6 in adenomyosis and endometriosis to further elucidate its role in uterine disorders.

PMID:39903727 | DOI:10.1371/journal.pone.0317136

PLoS One. 2025 Feb 4;20(2):e0318767. doi: 10.1371/journal.pone.0318767. eCollection 2025.

ABSTRACT

The frequent bending of shearer cables during operation often leads to mechanical fatigue, posing risks to equipment safety. Accurately predicting the mechanical properties of these cables under bending conditions is crucial for improving the reliability and service life of shearers. This paper proposes a shearer optical fiber cable mechanical characteristics prediction model based on Temporal Convolutional Network (TCN), Bidirectional Long Short-Term Memory (BiLSTM), and Squeeze-and-Excitation Attention (SEAttention), referred to as the TCN-BiLSTM-SEAttention model. This method leverages TCN's causal and dilated convolution operations to capture long-term sequential features, BiLSTM's bidirectional information processing to ensure the completeness of sequence information, and the SEAttention mechanism to assign adaptive weights to features, effectively enhancing the focus on key features. The model's performance is validated through comparisons with multiple other models, and the contributions of input features to the model's predictions are quantified using Shapley Additive Explanations (SHAP). By learning the stress variation patterns between the optical fiber, power conductor, and control conductor in the shearer cable, the model enables accurate prediction of the stress in other cable conductors based on optical fiber stress data. Experiments were conducted using a shearer optical fiber cable bending simulation dataset with traction speeds of 6 m/min, 8 m/min, and 10 m/min. The results show that, compared to other predictive models, the proposed model achieves reductions in Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) to 0.0002, 0.0159, and 0.0126, respectively, with the coefficient of determination (R2) increasing to 0.981. The maximum deviation between predicted and actual values is only 0.86%, demonstrating outstanding prediction accuracy. SHAP feature analysis reveals that the control conductor features have the most substantial influence on predictions, with a SHAP value of 0.095. The research shows that the TCN-BiLSTM-SEAttention model demonstrates outstanding predictive capability under complex operating conditions, providing a novel approach for improving cable management and equipment safety through optical fiber monitoring technology in the intelligent development of coal mines, highlighting the potential of deep learning in complex mechanical predictions.

PMID:39903714 | DOI:10.1371/journal.pone.0318767

Biol Reprod. 2025 Feb 4:ioaf023. doi: 10.1093/biolre/ioaf023. Online ahead of print.

ABSTRACT

The number and distribution of follicles in each growth stage provides a reliable readout of ovarian health and function. Leveraging techniques for three-dimensional imaging of ovaries in toto has the potential to uncover total, accurate ovarian follicle counts. Due to the size and holistic nature of these images, counting oocytes is time consuming and difficult. The advent of machine-learning algorithms has allowed for the development of ultra-fast, automated methods to analyze microscopy images. In recent years, these pipelines have become more accessible to non-specialists. We used these tools to create OoCount, a high-throughput, open-source method for automatic oocyte segmentation and classification from fluorescent 3D microscopy images of whole mouse ovaries using a deep-learning convolutional neural network (CNN) based approach. We developed a fast tissue-clearing and imaging protocol to obtain 3D images of whole mount mouse ovaries. Fluorescently labeled oocytes from 3D images were manually annotated in Napari to develop a training dataset. This dataset was used to retrain StarDist using a CNN within DL4MicEverywhere to automatically label all oocytes in the ovary. In a second phase, we utilize Accelerated Pixel and Object Classification, a Napari plugin, to sort oocytes into growth stages. Here, we provide an end-to-end pipeline for producing high-quality 3D images of mouse ovaries and obtaining follicle counts and staging. We demonstrate how to customize OoCount to fit images produced in any lab. Using OoCount, we obtain accurate oocyte counts from each growth stage in the perinatal and adult ovary, improving our ability to study ovarian function and fertility.

PMID:39903695 | DOI:10.1093/biolre/ioaf023

Eur J Prev Cardiol. 2025 Feb 4:zwaf038. doi: 10.1093/eurjpc/zwaf038. Online ahead of print.

ABSTRACT

AIMS: Cardiovascular disease (CVD) is a leading global cause of mortality. Environmental factors are increasingly recognized as influential determinants of cardiovascular health. Nevertheless, a finer-grained understanding of the effects of the built environment remains crucial for comprehending CVD. We sought to investigate the relationship between built environment features, including residential greenspace and sidewalks, and cardiovascular risk using street-level imagery and deep learning techniques.

METHODS: This study employed Google Street View (GSV) imagery and deep learning techniques to analyze built environment features around residences in relation to major adverse cardiovascular events (MACE) risk. Data from a Northeast Ohio cohort were utilized. Various covariates, including socioeconomic and environmental factors, were incorporated in Cox Proportional Hazards models.

RESULTS: Of 49,887 individuals included, 2,083 experienced MACE over a median follow-up of 26.86 months. Higher tree-sky index and sidewalk presence were associated with reduced MACE risk (HR: 0.95, 95% CI: 0.91-0.99, and HR: 0.91, 95% CI: 0.87-0.96, respectively), even after adjusting for demographic, socioeconomic, environmental, and clinical factors.

CONCLUSIONS: Visible vertical greenspace and sidewalks, as discerned from street-level images using deep learning, demonstrated potential associations with cardiovascular risk. This innovative approach highlights the potential of deep learning to analyze built environments at scale, offering new avenues for public health research. Future research is needed to validate these associations and better understand the underlying mechanisms.

PMID:39903569 | DOI:10.1093/eurjpc/zwaf038

Eur Radiol. 2025 Feb 4. doi: 10.1007/s00330-025-11376-9. Online ahead of print.

ABSTRACT

OBJECTIVE: To directly compare coronary arterial stenosis evaluations by hybrid-type iterative reconstruction (IR), model-based IR (MBIR), deep learning reconstruction (DLR), and high-resolution deep learning reconstruction (HR-DLR) on coronary computed tomography angiography (CCTA) in both in vitro and in vivo studies.

MATERIALS AND METHODS: For the in vitro study, a total of three-vessel tube phantoms with diameters of 3 mm, 4 mm, and 5 mm and with simulated non-calcified stepped stenosis plaques with degrees of 0%, 25%, 50%, and 75% stenosis were scanned with area-detector CT (ADCT) and ultra-high-resolution CT (UHR-CT). Then, ADCT data were reconstructed using all methods, although UHR-CT data were reconstructed with hybrid-type IR, MBIR, and DLR. For the in vivo study, patients who had undergone CCTA at ADCT were retrospectively selected, and each CCTA data set was reconstructed with all methods. To compare the image noise and measurement accuracy at each of the stenosis levels, image noise, and inner diameter were evaluated and statistically compared. To determine the effect of HR-DLR on CAD-RADS evaluation accuracy, the accuracy of CAD-RADS categorization of all CCTAs was compared by using McNemar's test.

RESULTS: The image noise of HR-DLR was significantly lower than that of others on ADCT and UHR-CT (p < 0.0001). At a 50% and 75% stenosis level for each phantom, hybrid-type IR showed a significantly larger mean difference on ADCT than did others (p < 0.05). At in vivo study, 31 patients were included. Accuracy on HR-DLR was significantly higher than that on hybrid-type IR, MBIR, or DLR (p < 0.0001).

CONCLUSION: HR-DLR is potentially superior for coronary arterial stenosis evaluations to hybrid-type IR, MBIR, or DLR shown on CCTA.

KEY POINTS: Question How do coronary arterial stenosis evaluations by hybrid-type IR, MBIR, DLR, and HR-DLR compare to coronary CT angiography? Findings HR-DLR showed significantly lower image noise and more accurate coronary artery disease reporting and data system (CAD-RADS) evaluation than others. Clinical relevance HR-DLR is potentially superior to other reconstruction methods for coronary arterial stenosis evaluations, as demonstrated by coronary CT angiography results on ADCT and as shown in both in vitro and in vivo studies.

PMID:39903239 | DOI:10.1007/s00330-025-11376-9