Abdom Radiol (NY). 2025 May 13. doi: 10.1007/s00261-025-04984-y. Online ahead of print.

ABSTRACT

OBJECTIVE: To evaluate the possibility of performing renal vessel reconstruction on non-enhanced CT images using deep learning models.

MATERIALS AND METHODS: 177 patients' CT scans in the non-enhanced phase, arterial phase and venous phase were chosen. These data were randomly divided into the training set (n = 120), validation set (n = 20) and test set (n = 37). In training set and validation set, a radiologist marked out the right renal arteries and veins on non-enhanced CT phase images using contrast phases as references. Trained deep learning models were tested and evaluated on the test set. A radiologist performed renal vessel reconstruction on the test set without the contrast phase reference, and the results were used for comparison. Reconstruction using the arterial phase and venous phase was used as the gold standard.

RESULTS: Without the contrast phase reference, both radiologist and model could accurately identify artery and vein main trunk. The accuracy was 91.9% vs. 97.3% (model vs. radiologist) in artery and 91.9% vs. 100% in vein, the difference was insignificant. The model had difficulty identify accessory arteries, the accuracy was significantly lower than radiologist (44.4% vs. 77.8%, p = 0.044). The model also had lower accuracy in accessory veins, but the difference was insignificant (64.3% vs. 85.7%, p = 0.094).

CONCLUSION: Deep learning models could accurately recognize the right renal artery and vein main trunk, and accuracy was comparable to that of radiologists. Although the current model still had difficulty recognizing small accessory vessels, further training and model optimization would solve these problems.

PMID:40358703 | DOI:10.1007/s00261-025-04984-y

Eur J Nucl Med Mol Imaging. 2025 May 13. doi: 10.1007/s00259-025-07321-z. Online ahead of print.

ABSTRACT

PURPOSE: In this study, we investigated the utility of a novel developed anti-CEACAM5 VHH for cancer diagnosis and its potential of being a targeting-moiety of VHH-drug conjugates for cancer therapy.

METHODS: Anti-CEACAM5 VHH (6B11) affinity and specific cellular binding was confirmed by ELISA, FACS and immunofluorescence in cancer cell lines with varying CEACAM5 expression levels. Intracellular penetration ability within tumor spheroids was tested with Oregon Green 488 labeled 6B11 (OG488-6B11). Biodistribution and binding specificity of 99mTc-radiolabeled 6B11 was tested in A549 CEACAM5 overexpressing (A549-CEA5-OV) and knockout (A549-CEA5-KO) tumor-bearing mice upon SPECT/CT imaging, γ-counting and autoradiography. The therapeutic efficacy of 6B11 and 6F8 (anti-CEACAM5 VHH with lower binding affinity) was tested by viability, wound healing and adhesion assays. To verify the potential of VHHs as a warhead for VHH-drug conjugation, an internalization assay with OG488 labeled VHH was performed.

RESULT: 6B11 demonstrated high binding affinity (EC50 0.5nM) and cellular binding. OG488-6B11 penetrated tumor spheroids completely at 24 h, while a conventional antibody was only visible at the spheroid periphery. SPECT imaging indicated higher uptake (p < 0.05) in A549-CEA5-OV tumors, resulting in increased tumor-to-blood ratios especially at 4 (2.0016 ± 1.1893, p = 0.035) and 24 (2.9371 ± 2.0683, p = 0.003) hpi compared to A549-CEA5-KO tumors at 4 (0.5640 ± 0.3576) and 24 (0.8051 ± 0.4351) hpi. 99mTc-6B11 was predominantly renally cleared. Autoradiography and immunohistochemistry confirmed these uptake patterns. 6B11 nor 6F8 did exhibit significant anti-cancer therapeutic efficacy in vitro. OG488-6B11 was effectively internalized and accumulated in cells in a time-dependent manner, to end up in the lysosomes.

CONCLUSION: The anti-CEACAM5 VHH 6B11 is a good candidate for SPECT-based cancer diagnosis and can be potentially used as targeting moiety in the development of VHH-based drug conjugates for cancer treatments.

PMID:40358697 | DOI:10.1007/s00259-025-07321-z

Int J Surg. 2025 May 12. doi: 10.1097/JS9.0000000000002486. Online ahead of print.

ABSTRACT

BACKGROUND: Given the current limited accuracy of imaging screening for Hepatic Echinococcosis (HCE) in under-resourced areas, the authors developed and validated a Multimodal Imaging system (HEAC) based on plain Computed Tomography (CT) combined with ultrasound for HCE screening in those areas.

METHODS: In this study, we developed a multimodal deep learning diagnostic system by integrating ultrasound and plain CT imaging data to differentiate hepatic echinococcosis, liver cysts, liver abscesses, and healthy liver conditions. We collected a dataset of 8979 cases spanning 18 years from eight hospitals in Xinjiang China, including both retrospective and prospective data. To enhance the robustness and generalization of the diagnostic model, after modeling CT and ultrasound images using EfficientNet3D and EfficientNet-B0, external and prospective tests were conducted, and the model's performance was compared with diagnoses made by experienced physicians.

RESULTS: Across internal and external test sets, the fused model of CT and ultrasound consistently outperformed the individual modality models and physician diagnoses. In the prospective test set from the same center, the fusion model achieved an accuracy of 0.816, sensitivity of 0.849, specificity of 0.942, and an AUC of 0.963, significantly exceeding physician performance (accuracy 0.900, sensitivity 0.800, specificity 0.933). The external test sets across seven other centers demonstrated similar results, with the fusion model achieving an overall accuracy of 0.849, sensitivity of 0.859, specificity of 0.942, and AUC of 0.961.

CONCLUSION: The multimodal deep learning diagnostic system that integrates CT and ultrasound significantly increases the diagnosis accuracy of HCE, liver cysts, and liver abscesses. It beats standard single-modal approaches and physician diagnoses by lowering misdiagnosis rates and increasing diagnostic reliability. It emphasizes the promise of multimodal imaging systems in tackling diagnostic issues in low-resource areas, opening the path for improved medical care accessibility and outcomes.

PMID:40358633 | DOI:10.1097/JS9.0000000000002486

Bioinformatics. 2025 May 13:btaf302. doi: 10.1093/bioinformatics/btaf302. Online ahead of print.

ABSTRACT

MOTIVATION: Advances in high throughput technologies enabled large access to various types of omics. Each omics provides a partial view of the underlying biological process. Integrating multiple omics layers would help have a more accurate diagnosis. However, the complexity of omics data requires approaches that can capture complex relationships. One way to accomplish this is by exploiting the known regulatory links between the different omics, which could help in constructing a better multimodal representation.

RESULTS: In this article, we propose CrossAttOmics, a new deep-learning architecture based on the cross-attention mechanism for multi-omics integration. Each modality is projected in a lower dimensional space with its specific encoder. Interactions between modalities with known regulatory links are computed in the feature representation space with cross-attention. The results of different experiments carried out in this paper show that our model can accurately predict the types of cancer by exploiting the interactions between multiple modalities. CrossAttOmics outperforms other methods when there are few paired training examples. Our approach can be combined with attribution methods like LRP to identify which interactions are the most important.

AVAILABILITY: The code is available at https://github.com/Sanofi-Public/CrossAttOmics and https://doi.org/10.5281/zenodo.15065928. TCGA data can be downloaded from the Genomic Data Commons Data Portal. CCLE data can be downloaded from the depmap portal.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PMID:40358524 | DOI:10.1093/bioinformatics/btaf302

Bioinformatics. 2025 May 13:btaf303. doi: 10.1093/bioinformatics/btaf303. Online ahead of print.

ABSTRACT

MOTIVATION: Chronic kidney disease (CKD) and Acute Kidney Injury (AKI) are prominent public health concerns affecting more than 15% of the global population. The ongoing development of spatially resolved transcriptomics (SRT) technologies presents a promising approach for discovering the spatial distribution patterns of gene expression within diseased tissues. However, existing computational tools are predominantly calibrated and designed on the ribbon-like structure of the brain cortex, presenting considerable computational obstacles in discerning highly heterogeneous mosaic-like tissue architectures in the kidney. Consequently, timely and cost-effective acquisition of annotation and interpretation in the kidney remains a challenge in exploring the cellular and morphological changes within renal tubules and their interstitial niches.

RESULTS: We present an empowered graph deep learning framework, REGNN (Relation Equivariant Graph Neural Networks), designed for SRT data analyses on heterogeneous tissue structures. To increase expressive power in the SRT lattice using graph modeling, REGNN integrates equivariance to handle n-dimensional symmetries of the spatial area, while additionally leveraging Positional Encoding to strengthen relative spatial relations of the nodes uniformly distributed in the lattice. Given the limited availability of well-labeled spatial data, this framework implements both graph autoencoder and graph self-supervised learning strategies. On heterogeneous samples from different kidney conditions, REGNN outperforms existing computational tools in identifying tissue architectures within the 10X Visium platform. This framework offers a powerful graph deep learning tool for investigating tissues within highly heterogeneous expression patterns and paves the way to pinpoint underlying pathological mechanisms that contribute to the progression of complex diseases.

AVAILABILITY: REGNN is publicly available at https://github.com/Mraina99/REGNN.

SUPPLEMENTARY INFORMATION: Found in the attached supplementary file 'SupplementaryFile_ManuscriptBioinformatics'.

PMID:40358510 | DOI:10.1093/bioinformatics/btaf303

Comput Methods Biomech Biomed Engin. 2025 May 13:1-25. doi: 10.1080/10255842.2025.2501634. Online ahead of print.

ABSTRACT

This study aims to introduce a novel, computationally lightweight feature extraction technique called Sliding Euclidean Pattern Quantization (SEPQ), which encodes local morphological patterns of ECG signals using Euclidean distance-based binary representations within sliding windows. The proposed SEPQ method was evaluated using two ECG datasets. The first dataset contained three labeled classes (CHF, ARR, and NSR), while the second included four classes (ventricular beats (VB), supraventricular beats (SVB), fusion beats (FB), and NSR). Extracted features were classified using several machine learning models, with LightGBM achieving the highest performance-over 99% accuracy on the first dataset and above 93% on the second. A convolutional neural network (CNN) model was also employed for comparative analysis, both on raw data and in a hybrid configuration with SEPQ, yielding moderate yet noteworthy performance. Experimental results confirm that SEPQ offers a robust, interpretable, and highly accurate solution for ECG signal classification.

PMID:40358468 | DOI:10.1080/10255842.2025.2501634

Radiology. 2025 May;315(2):e243463. doi: 10.1148/radiol.243463.

ABSTRACT

Background Forced vital capacity (FVC) is a standard measure of respiratory function in patients with amyotrophic lateral sclerosis (ALS) but has limitations, particularly for patients with bulbar impairment. Purpose To determine the value of deep learning-based chest CT-derived respiratory parameters in predicting ALS progression and survival. Materials and Methods This retrospective study included patients with ALS diagnosed between January 2010 and July 2023 who underwent chest CT at a tertiary hospital. Deep learning-based software was used to measure lung and respiratory muscle volume, normalized for height as the lung volume index (LVI) and respiratory muscle index (RMI). Differences in these parameters across King clinical stages were assessed using ordinal logistic regression. Tracheostomy-free survival was evaluated using Cox regression and time-dependent receiver operating characteristic analysis. Subgroup analysis was conducted for patients with bulbar impairment. In addition, a Gaussian process regressor model was developed to estimate FVC based on lung volume, respiratory muscle volume, age, and sex. Results A total of 261 patients were included in the study (mean age, 65.2 years ± 11.9 [SD]; 156 male patients). LVI and RMI decreased with increasing King stage (both P < .001). The high LVI and high RMI groups had better survival (both P < .001). After adjustment, LVI (hazard ratio [HR] = 0.998 [95% CI: 0.996, 1.000]; P = .021) and RMI (HR = 0.992 [95% CI: 0.988, 0.996]; P < .001) remained independent prognostic factors. In patients with bulbar impairment, LVI (HR = 0.998 [95% CI: 0.996, 1.000]; P = .029) and RMI (HR = 0.991 [95% CI: 0.987, 0.996]; P < .001) were independent prognostic factors. Time-dependent receiver operating characteristic curve analysis revealed no significant differences in survival prediction performance among LVI, RMI, and FVC. The Gaussian process regressor model estimated FVC with approximately 8% error. Conclusion The deep learning-derived CT metrics LVI and RMI reflected ALS stage, enabled FVC prediction, and supported assessment in patients with limited respiratory function. © RSNA, 2025 Supplemental material is available for this article. See also the editorial by Rahsepar and Bedayat in this issue.

PMID:40358443 | DOI:10.1148/radiol.243463

J Am Heart Assoc. 2025 May 13:e040641. doi: 10.1161/JAHA.124.040641. Online ahead of print.

ABSTRACT

BACKGROUND: Atrial fibrillation (AF) can cause a reduction in left ventricular ejection fraction (LVEF) that resolves rapidly upon restoration of sinus rhythm. We used artificial intelligence to understand (1) how often transient LVEF reduction during AF is from mismeasurement due to AF's beat-to-beat variability and (2) whether true transient AF-LVEF reduction has prognostic significance.

METHODS: In this observational study, we analyzed all patients at a large academic center with a transthoracic echocardiogram in AF and subsequent transthoracic echocardiogram in sinus rhythm within 90 days. We classified patients by their clinically reported LVEFs: no AF-LVEF reduction, transient AF-LVEF reduction that recovered after conversion to sinus rhythm, or persistent AF-LVEF reduction that did not recover. We evaluated how automated multicycle AF-LVEF measurement using a validated artificial intelligence algorithm affected AF-LVEF and reclassified patients. We used Fine-Gray hazard modeling to analyze 1-year heart failure hospitalization risk.

RESULTS: In 810 patients (mean age 74.1 years, 34.3% female), 459 (56.7%) had no reduced AF-LVEF, 71 (8.8%) had transient AF-LVEF reduction, and 280 (34.6%) had persistent AF-LVEF reduction. In the group with transient AF-LVEF reduction, LVEF increased by 19.5% (95% CI, 12.0%-22.1%) upon conversion to sinus rhythm. AI reassessment increased AF-LVEF by 8.2% (95% CI, 6.0%-10.4%), reclassifying 20 (28.2%) patients as no longer having reduced AF-LVEF. The group with transient AF-LVEF reduction, as determined by AI, had significantly higher 1-year heart failure hospitalization risk (hazard ratio, 2.28 [95% CI, 1.23-4.21], P=0.003).

CONCLUSION: Artificial intelligence may decrease misdiagnosis of reduced LVEF during AF and more accurately identify true transient AF-LVEF reduction, a potentially high-risk phenotype.

PMID:40357662 | DOI:10.1161/JAHA.124.040641

Clin Nucl Med. 2025 May 9. doi: 10.1097/RLU.0000000000005942. Online ahead of print.

ABSTRACT

PURPOSE: To evaluate the diagnostic performance of the PET Assisted Reporting System (PARS) in nasopharyngeal carcinoma (NPC) patients without distant metastasis, and to investigate the prognostic significance of the metabolic parameters.

PATIENTS AND METHODS: Eighty-three NPC patients who underwent pretreatment 18F-FDG PET/CT were retrospectively collected. First, the sensitivity, specificity, and accuracy of PARS for diagnosing malignant lesions were calculated, using histopathology as the gold standard. Next, metabolic parameters of the primary tumor were derived using both PARS and manual segmentation. The differences and consistency between the 2 methods were analyzed. Finally, the prognostic value of PET metabolic parameters was evaluated. Prognostic analysis of progression-free survival (PFS) and overall survival (OS) was conducted.

RESULTS: PARS demonstrated high patient-based accuracy (97.2%), sensitivity (88.9%), and specificity (97.4%), and 96.7%, 84.0%, and 96.9% based on lesions. Manual segmentation yielded higher metabolic tumor volume (MTV) and total lesion glycolysis (TLG) than PARS. Metabolic parameters from both methods were highly correlated and consistent. ROC analysis showed metabolic parameters exhibited differences in prognostic prediction, but generally performed well in predicting 3-year PFS and OS overall. MTV and age were independent prognostic factors; Cox proportional-hazards models incorporating them showed significant predictive improvements when combined. Kaplan-Meier analysis confirmed better prognosis in the low-risk group based on combined indicators (χ² = 42.25, P < 0.001; χ² = 20.44, P < 0.001).

CONCLUSIONS: Preliminary validation of PARS in NPC patients without distant metastasis shows high diagnostic sensitivity and accuracy for lesion identification and classification, and metabolic parameters correlate well with manual. MTV reflects prognosis, and its combination with age enhances prognostic prediction and risk stratification.

PMID:40357637 | DOI:10.1097/RLU.0000000000005942

Stud Health Technol Inform. 2025 May 12;326:147-151. doi: 10.3233/SHTI250257.

ABSTRACT

INTRODUCTION: Artificial Intelligence (AI) is playing an increasing role in advancing diagnostic processes and decision-making in healthcare. In the early detection of oral cancer and oral potentially malignant disorders (OPMDs), its role is still being explored. This paper evaluates advancements in AI applications for the early detection of oral cancer and OPMDs.

METHODS: A narrative umbrella review was performed on reviews that explicitly evaluated non-invasive diagnostic techniques combined with AI-modalities or machine learning techniques in the early detection of oral cancer and OPMDs.

RESULTS: Key findings of eight studies published between 2015 and 2024 demonstrate various AI-modalities and their diagnostic accuracy, accessibility and affordability, limitations and challenges and ethical and regulatory needs.

CONCLUSION: AI- and deep learning models hold promise in improving the early detection of oral cancer and OPMDs, offering high diagnostic accuracy that can significantly enhance patient outcomes. Challenges such as limited explainability and ethical concerns must be addressed to fully integrate these technologies into daily clinical practice.

PMID:40357619 | DOI:10.3233/SHTI250257

Circ Cardiovasc Imaging. 2025 May 13:e018188. doi: 10.1161/CIRCIMAGING.124.018188. Online ahead of print.

ABSTRACT

BACKGROUND: Computed tomography (CT) attenuation correction scans are an intrinsic part of positron emission tomography (PET) myocardial perfusion imaging using PET/CT, but anatomic information is rarely derived from these ultralow-dose CT scans. We aimed to assess the association between deep learning-derived cardiac chamber volumes (right atrial, right ventricular, left ventricular, and left atrial) and mass (left ventricular) from these scans with myocardial flow reserve and heart failure hospitalization.

METHODS: We included 18 079 patients with consecutive cardiac PET/CT from 6 sites. A deep learning model estimated cardiac chamber volumes and left ventricular mass from computed tomography attenuation correction imaging. Associations between deep learning-derived CT mass and volumes with heart failure hospitalization and reduced myocardial flow reserve were assessed in a multivariable analysis.

RESULTS: During a median follow-up of 4.3 years, 1721 (9.5%) patients experienced heart failure hospitalization. Patients with 3 or 4 abnormal chamber volumes were 7× more likely to be hospitalized for heart failure compared with patients with normal volumes. In adjusted analyses, left atrial volume (hazard ratio [HR], 1.25 [95% CI, 1.19-1.30]), right atrial volume (HR, 1.29 [95% CI, 1.23-1.35]), right ventricular volume (HR, 1.25 [95% CI, 1.20-1.31]), left ventricular volume (HR, 1.27 [95% CI, 1.23-1.35]), and left ventricular mass (HR, 1.25 [95% CI, 1.18-1.32]) were independently associated with heart failure hospitalization. In multivariable analyses, left atrial volume (odds ratio, 1.14 [95% CI, 1.0-1.19]) and ventricular mass (odds ratio, 1.12 [95% CI, 1.6-1.17]) were independent predictors of reduced myocardial flow reserve.

CONCLUSIONS: Deep learning-derived chamber volumes and left ventricular mass from computed tomography attenuation correction were predictive of heart failure hospitalization and reduced myocardial flow reserve in patients undergoing cardiac PET perfusion imaging. This anatomic data can be routinely reported along with other PET/CT parameters to improve risk prediction.

PMID:40357553 | DOI:10.1161/CIRCIMAGING.124.018188

Front Med (Lausanne). 2025 Apr 28;12:1530361. doi: 10.3389/fmed.2025.1530361. eCollection 2025.

ABSTRACT

OBJECTIVE: Accurate anterior visual pathway (AVP) segmentation is vital for clinical applications, but manual delineation is time-consuming and resource-intensive. We aim to explore the feasibility of automatic AVP segmentation and volume measurement in brain T1-weighted imaging (T1WI) using the 3D UX-Net deep-learning model.

METHODS: Clinical data and brain 3D T1WI from 119 adults were retrospectively collected. Two radiologists annotated the AVP course in each participant's images. The dataset was randomly divided into training (n = 89), validation (n = 15), and test sets (n = 15). A 3D UX-Net segmentation model was trained on the training data, with hyperparameters optimized using the validation set. Model accuracy was evaluated on the test set using Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and average symmetric surface distance (ASSD). The 3D UX-Net's performance was compared against 3D U-Net, Swin UNEt TRansformers (UNETR), UNETR++, and Swin Soft Mixture Transformer (Swin SMT). The AVP volume in the test set was calculated using the model's effective voxel volume, with volume difference (VD) assessing measurement accuracy. The average AVP volume across all subjects was derived from 3D UX-Net's automatic segmentation.

RESULTS: The 3D UX-Net achieved the highest DSC (0.893 ± 0.017), followed by Swin SMT (0.888 ± 0.018), 3D U-Net (0.875 ± 0.019), Swin UNETR (0.870 ± 0.017), and UNETR++ (0.861 ± 0.020). For surface distance metrics, 3D UX-Net demonstrated the lowest median ASSD (0.234 mm [0.188-0.273]). The VD of Swin SMT was significantly lower than that of 3D U-Net (p = 0.008), while no statistically significant differences were observed among other groups. All models exhibited identical HD95 (1 mm [1-1]). Automatic segmentation across all subjects yielded a mean AVP volume of 1446.78 ± 245.62 mm3, closely matching manual segmentations (VD = 0.068 ± 0.064). Significant sex-based volume differences were identified (p < 0.001), but no age correlation was observed.

CONCLUSION: We provide normative values for the automatic MRI measurement of the AVP in adults. The 3D UX-Net model based on brain T1WI achieves high accuracy in segmenting and measuring the volume of the AVP.

PMID:40357297 | PMC:PMC12066431 | DOI:10.3389/fmed.2025.1530361

Front Med (Lausanne). 2025 Apr 28;12:1564446. doi: 10.3389/fmed.2025.1564446. eCollection 2025.

ABSTRACT

BACKGROUND: Transperineal ultrasound (TPUS) is widely utilized for the evaluation of female stress urinary incontinence (SUI). However, the diagnostic accuracy of parameters related to urethral mobility and morphology remains limited and requires further optimization.

OBJECTIVE: This study aims to develop and validate an optimized deep learning (DL) model based on TPUS images to improve the precision and reliability of female SUI diagnosis.

METHODS: This retrospective study analyzed TPUS images from 464 women, including 200 patients with SUI and 264 controls, collected between 2020 and 2024. Three DL models (ResNet-50, ResNet-152, and DenseNet-121) were trained on resting-state and Valsalva-state images using an 8:2 training-to-testing split. Model performance was assessed using diagnostic metrics, including area under the curve (AUC), accuracy, sensitivity, and specificity. A TPUS-index model, constructed using measurement parameters assessing urethral mobility, was used for comparison. Finally, the best-performing DL model was selected to evaluate its diagnostic advantages over traditional methods.

RESULTS: Among the three developed DL models, DenseNet-121 demonstrated the highest diagnostic performance, achieving an AUC of 0.869, an accuracy of 0.87, a sensitivity of 0.872, a specificity of 0.761, a negative predictive value (NPV) of 0.788, and a positive predictive value (PPV) of 0.853. When compared to the TPUS-index model, the DenseNet-121 model exhibited significantly superior diagnostic performance in both the training set (z = -2.088, p = 0.018) and the testing set (z = -1.997, p = 0.046).

CONCLUSION: This study demonstrates the potential of DL models, particularly DenseNet-121, to enhance the diagnosis of female SUI using TPUS images, providing a reliable and consistent diagnostic tool for clinical practice.

PMID:40357276 | PMC:PMC12066636 | DOI:10.3389/fmed.2025.1564446

Front Med (Lausanne). 2025 Apr 28;12:1567119. doi: 10.3389/fmed.2025.1567119. eCollection 2025.

ABSTRACT

The early diagnosis and accurate classification of lung cancer have a critical impact on clinical treatment and patient survival. The rise of artificial intelligence technology has led to breakthroughs in medical image analysis. The Lung-PET-CT-Dx public dataset was used for the model training and evaluation. The performance of the You Only Look Once (YOLO) series of models in the lung CT image object detection task is compared in terms of algorithms, and different versions of YOLOv5, YOLOv8, YOLOv9, YOLOv10, and YOLOv11 are examined for lung cancer detection and classification. The experimental results indicate that the prediction results of YOLOv8 are better than those of the other YOLO versions, with a precision rate of 90.32% and a recall rate of 84.91%, which proves that the model can effectively assist physicians in lung cancer diagnosis and improve the accuracy of disease localization and identification.

PMID:40357272 | PMC:PMC12067791 | DOI:10.3389/fmed.2025.1567119

Front Med (Lausanne). 2025 Apr 28;12:1587307. doi: 10.3389/fmed.2025.1587307. eCollection 2025.

ABSTRACT

INTRODUCTION: Aortic stenosis (AS) is a valvular heart disease that obstructs normal blood flow from the left ventricle to the aorta due to pathological changes in the valve, leading to impaired cardiac function. Echocardiography is a key diagnostic tool for AS; however, its accuracy is influenced by inter-observer variability, operator experience, and image quality, which can result in misdiagnosis. Therefore, alternative methods are needed to assist healthcare professionals in achieving more accurate diagnoses.

METHODS: We proposed a deep learning model, RSMAS-Net, for the automated identification and diagnosis of AS using echocardiography. The model enhanced the ResNet50 backbone by replacing Stage 4 with Spatial and Channel Reconstruction Convolution (SCConv) and Multi-Dconv Head Transposed Attention (MDTA) modules, aiming to reduce redundant computations and improve feature extraction capabilities.

RESULTS: The proposed method was evaluated on the TMED-2 echocardiography dataset, achieving an accuracy of 94.67%, an F 1-score of 94.37%, and an AUC of 0.95 for AS identification. Additionally, the model achieved an AUC of 0.93 for AS severity classification on TMED-2. RSMAS-Net outperformed multiple baseline models in recall, precision, parameter efficiency, and inference time. It also achieved an AUC of 0.91 on the TMED-1 dataset.

CONCLUSION: RSMAS-Net effectively diagnoses and classifies the severity of AS in echocardiographic images. The integration of SCConv and MDTA modules enhances diagnostic accuracy while reducing model complexity compared to the original ResNet50 architecture. These results highlight the potential of RSMAS-Net in improving AS assessment and supporting clinical decision-making.

PMID:40357270 | PMC:PMC12066763 | DOI:10.3389/fmed.2025.1587307

BMJ Open Sport Exerc Med. 2025 Apr 15;11(2):e002388. doi: 10.1136/bmjsem-2024-002388. eCollection 2025.

ABSTRACT

BACKGROUND: The most talented young athletes often face challenges related to sports health problems (ie, injury and illness), largely due to inappropriate training, condensed competition schedules and high demands. Previous preventive measures in Norway have lacked successful integration into young athletes' routines, highlighting the need for a systematic approach to safeguarding their health.

OBJECTIVE: To document the development of Prep to be PRO, an educational module-based programme, designed to support the development and protect the health of young athletes enrolled in sports junior high schools and sports academy high schools. Prep to be PRO aims to empower athletes with the relevant knowledge and skills to prevent health problems.

METHODS: The development process, guided by the Translating Research into Injury Prevention Practice framework, involved extensive collaboration with school leaders, coaches and athletes. From June 2019 to June 2023, the process incorporated multidisciplinary input from more than 40 stakeholders, including health personnel, as well as experts in sports science, nutrition and sports psychology.

RESULTS: Prep to be PRO consists of 10 modules tailored for both sports Junior high schools and sports academy high schools. The modules cover a range of topics, including performance training, growth and maturation, load progression, recovery, total load, nutrition and sports psychology. The programme is athlete-centred, but coach-driven, including student-active approaches, collaboration, use of digital tools and deep learning. Prep to be PRO is anchored in the National High School Curriculum, ensuring relevance and alignment with educational standards. Specific competence goals and learning objectives from the curriculum are addressed and linked to each individual module.

CONCLUSIONS: This educational programme appears to be a notable step forward in the Norwegian sports school's approach. Specifically, it may enhance the focus on overall health, introduce an individualised approach and foster long-term athlete development. The integration into the national curriculum and the involvement of school staff in its delivery is expected to facilitate implementation. Future work will focus on the next phases of implementation, as systematic data collection from coaches and athletes, ongoing stakeholder engagement, continuous adaptation and support for educators to ensure fidelity and relevance. Updates and analyses from all evaluations will examine the programme's effectiveness. Long-term sustainability will be secured by organisational commitment, resource alignment and integrating the initiative into existing structures.

PMID:40357054 | PMC:PMC12067783 | DOI:10.1136/bmjsem-2024-002388

Oral (Basel). 2025 Mar;5(1):2. doi: 10.3390/oral5010002. Epub 2025 Jan 6.

ABSTRACT

Artificial intelligence (AI), particularly through machine learning (ML) and deep learning (DL) techniques such as convolutional neural networks (CNNs) and natural language processing (NLP), has shown remarkable promise in image analysis and clinical documentation in oral pathology. In order to explore the transformative potential of artificial intelligence (AI) in oral pathology, this review highlights key studies demonstrating current AI's improvement in oral pathology, such as detecting oral diseases accurately and streamlining diagnostic processes. However, several limitations, such as data quality, generalizability, legal and ethical considerations, financial constraints, and the need for paradigm shifts in practice, are critically examined. Addressing these challenges through collaborative efforts, robust validation, and strategic integration can pave the way for AI to revolutionize oral pathology, ultimately improving patient outcomes and advancing the field.

PMID:40357025 | PMC:PMC12068879 | DOI:10.3390/oral5010002

Eur Radiol. 2025 May 12. doi: 10.1007/s00330-025-11655-5. Online ahead of print.

ABSTRACT

OBJECTIVE: To develop a novel MRI-based deep learning (DL) diagnostic model, utilizing multicenter large-sample data, for the preoperative differentiation of sinonasal inverted papilloma (SIP) from SIP-transformed squamous cell carcinoma (SIP-SCC).

METHODS: This study included 568 patients from four centers with confirmed SIP (n = 421) and SIP-SCC (n = 147). Deep learning models were built using T1WI, T2WI, and CE-T1WI. A combined model was constructed by integrating these features through an attention mechanism. The diagnostic performance of radiologists, both with and without the model's assistance, was compared. Model performance was evaluated through receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA).

RESULTS: The combined model demonstrated superior performance in differentiating SIP from SIP-SCC, achieving AUCs of 0.954, 0.897, and 0.859 in the training, internal validation, and external validation cohorts, respectively. It showed optimal accuracy, stability, and clinical benefit, as confirmed by Brier scores and calibration curves. The diagnostic performance of radiologists, especially for less experienced ones, was significantly improved with model assistance.

CONCLUSIONS: The MRI-based deep learning model enhances the capability to predict malignant transformation of sinonasal inverted papilloma before surgery. By facilitating earlier diagnosis and promoting timely pathological examination or surgical intervention, this approach holds the potential to enhance patient prognosis.

KEY POINTS: Questions Sinonasal inverted papilloma (SIP) is prone to malignant transformation locally, leading to poor prognosis; current diagnostic methods are invasive and inaccurate, necessitating effective preoperative differentiation. Findings The MRI-based deep learning model accurately diagnoses malignant transformations of SIP, enabling junior radiologists to achieve greater clinical benefits with the assistance of the model. Clinical relevance A novel MRI-based deep learning model enhances the capability of preoperative diagnosis of malignant transformation in sinonasal inverted papilloma, providing a non-invasive tool for personalized treatment planning.

PMID:40355636 | DOI:10.1007/s00330-025-11655-5

Sci Rep. 2025 May 12;15(1):16418. doi: 10.1038/s41598-025-94127-6.

ABSTRACT

Detecting and classifying arrhythmias is essential in diagnosing cardiovascular diseases. However, current deep learning-based classification methods often encounter difficulties in effectively integrating both the morphological and temporal features of Electrocardiograms (ECGs). To address this challenge, we propose a Convolutional Neural Network (CNN) that incorporates mixed scales and hierarchical features combined with the Lead Encoder Attention (LEA) mechanism for multi-lead ECG classification. We validated the performance of our proposed method using the intrapatient approach of the MIT-BIH Arrhythmia (MIT-BIH-AR) Database and the interpatient approach of the Chinese Cardiovascular Disease Database (CCDD). Our model achieves an Accuracy (Acc) of 99.5% for the classification of normal and abnormal heartbeats in the MIT-BIH-AR database. Our method achieves a TPR95 (NPV under the condition of True Positive Rate being equal to 95 percent) of 78.5% and an Acc of 88.5% when classifying normal and abnormal ECG records from over 150,000 ECG records in the CCDD. The cross-dataset experimental results also confirm the model's strong generalization capability.

PMID:40355498 | DOI:10.1038/s41598-025-94127-6

Sci Rep. 2025 May 12;15(1):16392. doi: 10.1038/s41598-025-01015-0.

ABSTRACT

Epilepsy is a neurological disorder characterized by recurrent seizures caused by excessive electrical discharges in brain cells, posing significant diagnostic and therapeutic challenges. Dynamic brain network analysis via electroencephalography (EEG) has emerged as a powerful tool for capturing transient functional connectivity changes, offering advantages over static networks. In this study, we propose a Dynamic Temporal-Spatial Graph Attention Network (DTS-GAN) to address the limitations of fixed-topology graph models in analysing time-varying brain networks. By integrating graph signal processing with a hybrid deep learning framework, DTS-GAN collaboratively extracts spatiotemporal features through two key modules: an LSTM-based temporal encoder to model long-term dependencies in EEG sequences, and a dynamic graph attention network with probabilistic Gaussian connectivity, enabling adaptive learning of transient functional interactions across electrode nodes. Experiments on the TUSZ dataset demonstrate that DTS-GAN achieves 89-91% accuracy and a weighted F1-score of 87-91% in classifying seven seizure types, significantly outperforming baseline models. The multi-head attention mechanism and dynamic graph generation strategy effectively resolve the temporal variability of functional connectivity. These results highlight the potential of DTS-GAN in providing precise and automated seizure detection, serving as a robust tool for clinical EEG analysis.

PMID:40355495 | DOI:10.1038/s41598-025-01015-0