JAMA Netw Open. 2025 Mar 3;8(3):e250984. doi: 10.1001/jamanetworkopen.2025.0984.

ABSTRACT

IMPORTANCE: While prospective studies have investigated the accuracy of artificial intelligence (AI) for detection of diabetic retinopathy (DR) and diabetic macular edema (DME), to date, little published data exist on the clinical performance of these algorithms.

OBJECTIVE: To evaluate the clinical performance of an automated retinal disease assessment (ARDA) algorithm in the postdeployment setting at Aravind Eye Hospital in India.

DESIGN, SETTING, AND PARTICIPANTS: This cross-sectional analysis involved an approximate 1% sample of fundus photographs from patients screened using ARDA. Images were graded via adjudication by US ophthalmologists for DR and DME, and ARDA's output was compared against the adjudicated grades at 45 sites in Southern India. Patients were randomly selected between January 1, 2019, and July 31, 2023.

MAIN OUTCOMES AND MEASURES: Primary analyses were the sensitivity and specificity of ARDA for severe nonproliferative DR (NPDR) or proliferative DR (PDR). Secondary analyses focused on sensitivity and specificity for sight-threatening DR (STDR) (DME or severe NPDR or PDR).

RESULTS: Among the 4537 patients with 4537 images with adjudicated grades, mean (SD) age was 55.2 (11.9) years and 2272 (50.1%) were male. Among the 3941 patients with gradable photographs, 683 (17.3%) had any DR, 146 (3.7%) had severe NPDR or PDR, 109 (2.8%) had PDR, and 398 (10.1%) had STDR. ARDA's sensitivity and specificity for severe NPDR or PDR were 97.0% (95% CI, 92.6%-99.2%) and 96.4% (95% CI, 95.7%-97.0%), respectively. Positive predictive value (PPV) was 50.7% and negative predictive value (NPV) was 99.9%. The clinically important miss rate for severe NPDR or PDR was 0% (eg, some patients with severe NPDR or PDR were interpreted as having moderate DR and referred to clinic). ARDA's sensitivity for STDR was 95.9% (95% CI, 93.0%-97.4%) and specificity was 94.9% (95% CI, 94.1%-95.7%); PPV and NPV were 67.9% and 99.5%, respectively.

CONCLUSIONS AND RELEVANCE: In this cross-sectional study investigating the clinical performance of ARDA, sensitivity and specificity for severe NPDR and PDR exceeded 96% and caught 100% of patients with severe NPDR and PDR for ophthalmology referral. This preliminary large-scale postmarketing report of the performance of ARDA after screening 600 000 patients in India underscores the importance of monitoring and publication an algorithm's clinical performance, consistent with recommendations by regulatory bodies.

PMID:40105843 | DOI:10.1001/jamanetworkopen.2025.0984

J Chem Theory Comput. 2025 Mar 19. doi: 10.1021/acs.jctc.4c01136. Online ahead of print.

ABSTRACT

A major challenge for many rare-event sampling strategies is the identification of progress coordinates that capture the slowest relevant motions. Machine-learning methods that can identify progress coordinates in an unsupervised manner have therefore been of great interest to the simulation community. Here, we developed a general method for identifying progress coordinates "on-the-fly" during weighted ensemble (WE) rare-event sampling via deep learning (DL) of outliers among sampled conformations. Our method identifies outliers in a latent space model of the system's sampled conformations that is periodically trained using a convolutional variational autoencoder. As a proof of principle, we applied our DL-enhanced WE method to simulate the NTL9 protein folding process. To enable rapid tests, our simulations propagated discrete-state synthetic molecular dynamics trajectories using a generative, fine-grained Markov state model. Results revealed that our on-the-fly DL of outliers enhanced the efficiency of WE by >3-fold in estimating the folding rate constant. Our efforts are a significant step forward in the unsupervised learning of slow coordinates during rare event sampling.

PMID:40105797 | DOI:10.1021/acs.jctc.4c01136

Technol Health Care. 2025 Mar 19:9287329241302736. doi: 10.1177/09287329241302736. Online ahead of print.

ABSTRACT

Ovarian cancer is a highly prevalent cancer among women; However, it remains difficult to find effective pharmacological solutions to treat this deadly disease. However, early detection can significantly increase life expectancy. To address this issue, a predictive model for early diagnosis of ovarian cancer was developed by applying statistical techniques and machine learning models to clinical data from 349 patients. A hybrid evolutionary deep learning model was proposed by integrating genetic and histopathological imaging modalities within a multimodal fusion framework. Machine learning pipelines have been built using feature selection and dilution approaches to identify the most relevant genes for disease classification. A comparison was performed between the UNeT and transformer models for semantic segmentation, leading to the development of an optimized fuzzy C-means clustering algorithm (FCM-NPOA-PM-UI) for the classification of gynecological abdominopelvic tumors. Performing better than individual classifiers and other machine learning methods, the suggested ensemble model achieved an average accuracy of 98.96%, precision of 97.44%, and F1 score of 98.7%. With average Dice scores of 0.98 and 0.97 for positive tumors and 0.99 and 0.98 for malignant tumors, the Transformer model performed better in segmentation than the UNeT model. Additionally, we observed a 92.8% increase in accuracy when combining five machine learning models with biomarker data: random forest, logistic regression, SVM, decision tree, and CNN. These results demonstrate that the hybrid model significantly improves the accuracy and efficiency of ovarian cancer detection and classification, offering superior performance compared to traditional methods and individual classifiers.

PMID:40105378 | DOI:10.1177/09287329241302736

J Chem Inf Model. 2025 Mar 19. doi: 10.1021/acs.jcim.5c00199. Online ahead of print.

ABSTRACT

Cell-penetrating peptides (CPPs) are usually short oligopeptides with 5-30 amino acid residues. CPPs have been proven as important drug delivery vehicles into cells through different mechanisms, demonstrating their potential as therapeutic candidates. However, experimental screening and synthesis of CPPs could be time-consuming and expensive. Recently, numerous attempts have been made to develop computational methods as a cost-effective way for screening a number of potential CPP candidates. Despite significant advancements, current methods exhibit limited feature representation capabilities, thereby constraining the potential for further performance enhancements. In this study, we developed a deep learning framework called CPPCGM, which uses protein language models (PLMs) to identify and generate novel CPPs. There are two separate blocks in this framework: CPPClassifier and CPPGenerator. The former utilizes three pretrained models for simple voting, thereby accurately categorizing CPPs and non-CPPs. The latter, similar to a generative adversarial network, including a discriminator and a generator, generates peptides that are not present in the training data set. Our proposed CPPCGM has achieved remarkably high Matthews correlation coefficient scores of 0.876, 0.923, and 0.664 on three data sets based on the classification results. Compared with the state-of-the-art methods, the performance of our method is significantly improved. The results also demonstrated the generating potential of CPPCGM through qualitative and quantitative evaluation of the generated samples. Significantly, using PLM-based methods can optimize peptides for biochemical functions, benefiting drug delivery and biomedical applications. Materials related are publicly available at https://github.com/QiufenChen/CPPCGM.

PMID:40105337 | DOI:10.1021/acs.jcim.5c00199

Eur J Prosthodont Restor Dent. 2025 Mar 13. doi: 10.1922/EJPRD_2801Hassan09. Online ahead of print.

ABSTRACT

OBJECTIVES: This study aimed to develop an artificial intelligence system for automated classification of partially edentulous arches from panoramic radiographs using the Kennedy classification system and Applegate's rules, alongside identifying existing teeth for automated reporting.

METHODS: From 5261 anonymized digital panoramic radiographs collected from publicly available datasets, 1875 high-quality images were selected and divided into training (80%), validation (10%), and testing (10%) sets. Teeth were manually annotated on the Roboflow platform following the Universal Numbering System. To enhance model robustness, data augmentation techniques were applied, expanding the dataset to 2398 images. For tooth detection, a YOLOv8s deep learning model was trained for 80 epochs (batch size: 16, learning rate: 0.01). Performance was evaluated using precision, recall, F1 score, and mean average precision. Detected teeth were used to classify partially edentulous areas based on the Kennedy system. Modification areas were identified by analyzing detected and missing teeth, measuring bounded distances in millimetres, and classifying free-end saddle gaps.

RESULTS: The YOLOv8s model achieved a mean average precision (mAP50) of 98.1% for tooth identification, with precision and recall of 95.7% and 95.8%, respectively. For Kennedy classification, the model demonstrated precision of 0.962, recall of 0.931, and an F1-score of 0.939 across maxillary and mandibular arches.

CONCLUSIONS: The high accuracy and efficiency of this AI-driven approach can standardize classification, reduce diagnostic variability, and alleviate the workload for dental professionals, enabling seamless integration into clinical practice.

CLINICAL RELEVANCE: This AI system provides a consistent, accurate, and reliable method for classifying partially edentulous arches from panoramic radiographs, reducing manual assessment variability, easing practitioner workload, and enabling large-scale analysis of partial edentulism prevalence.

PMID:40105321 | DOI:10.1922/EJPRD_2801Hassan09

mSystems. 2025 Mar 19:e0018325. doi: 10.1128/msystems.00183-25. Online ahead of print.

ABSTRACT

In the field of synthetic biology, the engineering of synthetic promoters that outperform their natural counterparts is of paramount importance, which can optimize the expression of exogenous genes, enhance the efficiency of metabolic pathways, and possess substantial commercial value. Research indicates that some synthetic promoters have higher transcriptional activity compared to strong natural promoters. However, with the exponential increase in complexity due to the 4n potential combinations in a promoter sequence of length n, identifying effective synthetic promoters remains a formidable challenge. Deep learning models, by adaptively learning from extensive data sets, have become instrumental in analyzing biological data. This study introduces a diffusion model-based approach for designing promoters viable in model bacteria such as Escherichia coli and cyanobacteria. This model proficiently assimilates and utilizes inherent biological features from natural promoter sequences to engineer synthetic variants. Additionally, we employed a transformer model to evaluate the efficacy of these synthetic promoters, aiming at screening those with high performance. The experimental findings suggest that the synthetic promoters by the diffusion model not only share key biological features with their natural counterparts but also demonstrate greater similarity to natural promoters than those generated by a variational autoencoder. In predicting promoter strength, the transformer model demonstrated improved performance over the convolutional neural network. Finally, we developed an integrated platform for generating promoters and predicting their strength.

IMPORTANCE: We demonstrated that diffusion models are superior in accomplishing the promoter synthesis task compared to other state-of-the-art deep learning models. The effectiveness of our method was validated using data sets of Escherichia coli and cyanobacteria promoters, showing more stable and prompt convergence and more natural-like promoters than the variational autoencoder model. We extracted sequence information, dimer information, and position information from promoters and combined them with a transformer model to predict promoter strength. Our prediction results were more accurate than those obtained with a convolutional neural network model. Our in silico experiments systematically introduced mutations in promoter sequences and explored their contribution to promoter strength, highlighting the depth of learning in our model.

PMID:40105319 | DOI:10.1128/msystems.00183-25

Nanoscale. 2025 Mar 19. doi: 10.1039/d4nr05031b. Online ahead of print.

ABSTRACT

Bismuth selenide (Bi2Se3) has emerged as a promising material for high-performance photodetectors due to its wideband spectral response, strong in-plane anisotropy, narrow bandgap, high absorption coefficient, and carrier mobility. However, inherent defects and states in Bi2Se3-based devices reduce optical conversion efficiency and stability. To address these challenges, we report the design and preparation of Bi2Se2.33S0.67 nanowires by a facile chemical vapor transport method. The individual Bi2Se2.33S0.67 nanowire photodetectors exhibit remarkable photoresponse over a broadband wavelength region ranging from ultraviolet C (254 nm) to near-infrared (1064 nm) with a low dark current of 0.015 nA and the measured maximum photoresponsivity of 2.52 A W-1 at 532 nm, together with a detectivity of around 5.2 × 1011 Jones. Furthermore, the photoresponse of photodetectors exhibits polarization angle sensitivity within a broadband range of 355 to 808 nm. The structural anisotropy of the Bi2Se2.33S0.67 crystal leads to a maximum dichroic ratio of about 1.8 at 355 nm. Additionally, cat images produced by this device further demonstrate the potential of the high-performance devices, and the effectiveness of photodetectors in deep learning image recognition validates their wide-spectrum, high-responsivity, and superior polarization-sensitive detection capabilities.

PMID:40105281 | DOI:10.1039/d4nr05031b

Technol Health Care. 2025 Mar;33(2):1099-1120. doi: 10.1177/09287329241296438. Epub 2024 Dec 4.

ABSTRACT

BackgroundBreast cancer results from an uncontrolled growth of breast tissue. Many methods of diagnosis are using multi-omics data to better understand the complexity of breast cancer.ObjectiveThe new strategy laid out in this work, called "Hybrid-OmniSeq," is a deep learning-based multi-omics data analysis technology that uses molecular subtypes of breast cancer gene to increase the precision and effectiveness of breast cancer diagnosis.MethodFor preprocessing, the BC-VM procedure is utilized, and for molecular subtype analysis, the BC-MSA procedure is utilized. The implementation of Deep Neural Network (DNN) technology in conjunction with Sequential Forward Floating Selection (SFFS) and Truncated Singular Value Decomposition (TSVD) entropy enable adaptive learning from multi-omics gene data. Five machine learning classifiers are used for classification purpose. Hybrid-OmniSeq uses a variety of machine learning classifiers in a thorough analytical process to achieve remarkable diagnostic accuracy. Deep Learning-based multi-omics sequential approach was evaluated using METABRIC RNA-seq data sets of intrinsic subtypes of breast cancer.ResultsAccording to test results, Logistic Regression (LR) had ER (Estrogen Receptor) status values of 94.51%, ER status values of 96.33%, and HER2 (Human Epidermal growth factor Receptor) status values of 92.3%; Random Forest (RF) had ER status values of 93.77%, ER status values of 95.23%, and HER2 status values of 93.4%.ConclusionLR and RF increase the cancer detection accuracy for all subtypes when compared to alternative machine learning classifiers or the majority voting method, providing a comprehensive understanding of the underlying causes of breast cancer.

PMID:40105178 | DOI:10.1177/09287329241296438

Technol Health Care. 2025 Mar;33(2):989-999. doi: 10.1177/09287329241297900. Epub 2024 Nov 20.

ABSTRACT

BackgroundDeep learning excels at processing raw data because it automatically extracts and classifies high-level features. Despite biology's low popularity in data analysis, incorporating computer technology can improve biological research.ObjectiveTo create a deep learning model that can identify nucleosomes from nucleotide sequences and to show that simpler models outperform more complicated ones in solving biological challenges.MethodsA classifier was created utilising deep learning and machine learning approaches. The final model consists of two convolutional layers, one max pooling layer, two fully connected layers, and a dropout regularisation layer. This structure was chosen on the basis of the 'less is frequently more' approach, which emphasises simple design without large hidden layers.ResultsExperimental results show that deep learning methods, specifically deep neural networks, outperform typical machine learning algorithms for recognising nucleosomes. The simplified network architecture proved suitable without the requirement for numerous hidden neurons, resulting in effective network performance.ConclusionThis study demonstrates that machine learning and other computational techniques may streamline and expedite the resolution of biological issues. The model helps identify nucleosomes and can be used in future research or labs. This study discusses the challenges of understanding and addressing simple biological problems with sophisticated computer technology and offers practical solutions for academic and economic sectors.

PMID:40105177 | DOI:10.1177/09287329241297900

Front Digit Health. 2025 Mar 4;7:1550407. doi: 10.3389/fdgth.2025.1550407. eCollection 2025.

ABSTRACT

Early diagnosis and accurate prognosis play a pivotal role in the clinical management of cancer and in preventing cancer-related mortalities. The burgeoning population of Asia in general and South Asian countries like India in particular pose significant challenges to the healthcare system. Regrettably, the demand for healthcare services in India far exceeds the available resources, resulting in overcrowded hospitals, prolonged wait times, and inadequate facilities. The scarcity of trained manpower in rural settings, lack of awareness and low penetrance of screening programs further compounded the problem. Artificial Intelligence (AI), driven by advancements in machine learning, deep learning, and natural language processing, can profoundly transform the underlying shortcomings in the healthcare industry, more for populous nations like India. With about 1.4 million cancer cases reported annually and 0.9 million deaths, India has a significant cancer burden that surpassed several nations. Further, India's diverse and large ethnic population is a data goldmine for healthcare research. Under these circumstances, AI-assisted technology, coupled with digital health solutions, could support effective oncology care and reduce the economic burden of GDP loss in terms of years of potential productive life lost (YPPLL) due to India's stupendous cancer burden. This review explores different aspects of cancer management, such as prevention, diagnosis, precision treatment, prognosis, and drug discovery, where AI has demonstrated promising clinical results. By harnessing the capabilities of AI in oncology research, healthcare professionals can enhance their ability to diagnose cancers at earlier stages, leading to more effective treatments and improved patient outcomes. With continued research and development, AI and digital health can play a transformative role in mitigating the challenges posed by the growing population and advancing the fight against cancer in India. Moreover, AI-driven technologies can assist in tailoring personalized treatment plans, optimizing therapeutic strategies, and supporting oncologists in making well-informed decisions. However, it is essential to ensure responsible implementation and address potential ethical and privacy concerns associated with using AI in healthcare.

PMID:40103737 | PMC:PMC11913822 | DOI:10.3389/fdgth.2025.1550407

Dentomaxillofac Radiol. 2025 Mar 18:twaf017. doi: 10.1093/dmfr/twaf017. Online ahead of print.

ABSTRACT

OBJECTIVES: Temporomandibular joint disorder (TMD) patients experience a variety of clinical symptoms, and magnetic resonance imaging (MRI) is the most effective tool for diagnosing temporomandibular joint (TMJ) disc displacement. This study aimed to develop a transformer-based deep learning model to generate T2-weighted (T2w) images from proton density-weighted (PDw) images, reducing MRI scan time for TMD patients.

METHODS: A dataset of 7,226 images from 178 patients who underwent TMJ MRI examinations was used. The proposed model employed a generative adversarial network framework with a TransUNet architecture as the generator for image translation. Additionally, a disc segmentation decoder was integrated to improve image quality in the TMJ disc region. The model performance was evaluated using metrics such as the structural similarity index measure (SSIM), learned perceptual image patch similarity (LPIPS), and Fréchet inception distance (FID). Three experienced oral radiologists also performed a qualitative assessment through the mean opinion score (MOS).

RESULTS: The model demonstrated high performance in generating T2w images from PDw images, achieving average SSIM, LPIPS, and FID values of 82.28%, 2.46, and 23.85, respectively, in the disc region. The model also obtained an average MOS score of 4.58, surpassing other models. Additionally, the model showed robust segmentation capabilities for the TMJ disc.

CONCLUSION: The proposed model using the transformer, complemented by an integrated disc segmentation task, demonstrated strong performance in MR image generation, both quantitatively and qualitatively. This suggests its potential clinical significance in reducing MRI scan times for TMD patients while maintaining high image quality.

PMID:40104864 | DOI:10.1093/dmfr/twaf017

NAR Genom Bioinform. 2025 Mar 18;7(1):lqaf021. doi: 10.1093/nargab/lqaf021. eCollection 2025 Mar.

ABSTRACT

The structural prediction of biomolecules via computational methods complements the often involved wet-lab experiments. Unlike protein structure prediction, RNA structure prediction remains a significant challenge in bioinformatics, primarily due to the scarcity of annotated RNA structure data and its varying quality. Many methods have used this limited data to train deep learning models but redundancy, data leakage and bad data quality hampers their performance. In this work, we present NucleoSeeker, a tool designed to curate high-quality, tailored datasets from the Protein Data Bank (PDB) database. It is a unified framework that combines multiple tools and streamlines an otherwise complicated process of data curation. It offers multiple filters at structure, sequence, and annotation levels, giving researchers full control over data curation. Further, we present several use cases. In particular, we demonstrate how NucleoSeeker allows the creation of a nonredundant RNA structure dataset to assess AlphaFold3's performance for RNA structure prediction. This demonstrates NucleoSeeker's effectiveness in curating valuable nonredundant tailored datasets to both train novel and judge existing methods. NucleoSeeker is very easy to use, highly flexible, and can significantly increase the quality of RNA structure datasets.

PMID:40104673 | PMC:PMC11915511 | DOI:10.1093/nargab/lqaf021

J Pathol Inform. 2025 Feb 4;17:100422. doi: 10.1016/j.jpi.2025.100422. eCollection 2025 Apr.

ABSTRACT

Meningiomas, the most prevalent tumors of the central nervous system, can have overlapping histopathological features with solitary fibrous tumors (SFT), presenting a significant diagnostic challenge. Accurate differentiation between these two diagnoses is crucial for optimal medical management. Currently, immunohistochemistry and molecular techniques are the methods of choice for distinguishing between them; however, these techniques are expensive and not universally available. In this article, we propose a rotational and scale-invariant deep learning framework to enable accurate discrimination between these two tumor types. The proposed framework employs a novel architecture of conical transformers to capture both global and local imaging markers from whole-slide images, accommodating variations across different magnification scales. A weighted majority voting schema is utilized to combine individual scale decisions, ultimately producing a complementary and more accurate diagnostic outcome. A dataset comprising 92 patients (46 with meningioma and 46 with SFT) was used for evaluation. The experimental results demonstrate robust performance across different validation methods. In train-test evaluation, the model achieved 92.27% accuracy, 87.77% sensitivity, 97.55% specificity, and 92.46% F1-score. Performance further improved in 4-fold cross-validation, achieving 94.68% accuracy, 96.05% sensitivity, 93.11% specificity, and 95.07% F1-score. These findings highlight the potential of AI-based diagnostic approaches for precise differentiation between meningioma and SFT, paving the way for innovative diagnostic tools in pathology.

PMID:40104410 | PMC:PMC11914819 | DOI:10.1016/j.jpi.2025.100422

Turk J Med Sci. 2023 Oct 16;55(1):161-177. doi: 10.55730/1300-0144.5954. eCollection 2025.

ABSTRACT

BACKGROUND/AIM: Skin lesions are commonly diagnosed and classified using dermoscopic images. There are many artifacts visible in dermoscopic images, including hair strands, noise, bubbles, blood vessels, poor illumination, and moles. These artifacts can obscure crucial information about lesions, which limits the ability to diagnose lesions automatically. This study investigated how hair and noise artifacts in lesion images affect classifier performance and how they can be removed to improve diagnostic accuracy.

MATERIALS AND METHODS: A synthetic dataset created using hair simulation and noise simulation was used in conjunction with the HAM10000 benchmark dataset. Moreover, integrated convolutional neural networks (CNNs) were proposed for removing hair artifacts using hair inpainting and classification of refined dehaired images, called integrated hair removal (IHR), and for removing noise artifacts using nonlocal mean denoising and classification of refined denoised images, called integrated noise removal (INR).

RESULTS: Five deep learning models were used for the classification: ResNet50, DenseNet121, ResNet152, VGG16, and VGG19. The proposed IHR-DenseNet121, IHR-ResNet50, and IHR-ResNet152 achieved 2.3%, 1.78%, and 1.89% higher accuracy than DenseNet121, ResNet50, and ResNet152, respectively, in removing hairs. The proposed INR-DenseNet121, INR-ResNet50, and INR-VGG19 achieved 1.41%, 2.39%, and 18.4% higher accuracy than DenseNet121, ResNet50, and VGG19, respectively, in removing noise.

CONCLUSION: A significant proportion of pixels within lesion areas are influenced by hair and noise, resulting in reduced classification accuracy. However, the proposed CNNs based on IHR and INR exhibit notably improved performance when restoring pixels affected by hair and noise. The performance outcomes of this proposed approach surpass those of existing methods.

PMID:40104314 | PMC:PMC11913500 | DOI:10.55730/1300-0144.5954

Turk J Med Sci. 2024 Aug 6;55(1):140-151. doi: 10.55730/1300-0144.5952. eCollection 2025.

ABSTRACT

BACKGROUND/AIM: Motor neuron disease (MND) is a devastating neuron ailment that affects the motor neurons that regulate muscular voluntary actions. It is a rare disorder that gradually destroys aspects of neurological function. In general, MND arises as a result of a combination of natural, behavioral, and genetic influences. However, early detection of MND is a challenging task and manual identification is time-consuming. To overcome this, a novel deep learning-based duple feature extraction framework is proposed for the early detection of MND.

MATERIALS AND METHODS: Diffusion tensor imaging tractography (DTI) images were initially analyzed for color and textural features using dual feature extraction. Local binary pattern (LBP)-based methods were used to extract textural data from images by examining nearby pixel values. A color information feature was then added to the LBP-based feature during the classification phase for extracting color features. A flattened image was then fed into the MONDNet for classifying normal and abnormal cases of MND based on color and texture features.

RESULTS: The proposed deep MONDNet is suitable because it achieved a detection rate of 99.66% and can identify MND in its early stages.

CONCLUSION: The proposed mobile net model achieved an overall F1 score of 13.26%, 6.15%, 5.56%, and 5.96% compared to the BPNN, CNN, SVM-RFE, and MLP algorithms, respectively.

PMID:40104302 | PMC:PMC11913516 | DOI:10.55730/1300-0144.5952

Phys Imaging Radiat Oncol. 2025 Feb 21;33:100736. doi: 10.1016/j.phro.2025.100736. eCollection 2025 Jan.

ABSTRACT

BACKGROUND AND PURPOSE: Delineation of target volumes (TVs) and organs at risk (OARs) is a resource intensive process in lung radiation therapy and, despite the introduction of some auto-contouring, inter-observer variability remains a challenge. Deep learning algorithms may prove an efficient alternative and this review aims to map the evidence base on the use of deep learning algorithms for TV and OAR delineation in the radiation therapy planning process for lung cancer patients.

MATERIALS AND METHODS: A literature search identified studies relating to deep learning. Manual contouring and deep learning auto-contours were evaluated against one another for accuracy, inter-observer variability, contouring time and dose-volume effects. A total of 40 studies were included for review.

RESULTS: Thirty nine out of 40 studies investigated the accuracy of deep learning auto-contours and determined that they were of a comparable accuracy to manual contours. Inter-observer variability outcomes were heterogeneous in the seven relevant studies identified. Twenty-four studies analysed the time saving associated with deep learning auto-contours and reported a significant time reduction in comparison to manual contours. The eight studies that conducted a dose-volume metric evaluation on deep learning auto-contours identified negligible effect on treatment plans.

CONCLUSION: The accuracy and time-saving capacity of deep learning auto-contours in comparison to manual contours has been extensively studied. However, additional research is required in the areas of inter-observer variability and dose-volume metric evaluation to further substantiate its clinical use.

PMID:40104215 | PMC:PMC11914827 | DOI:10.1016/j.phro.2025.100736

Ophthalmol Sci. 2025 Jan 28;5(3):100719. doi: 10.1016/j.xops.2025.100719. eCollection 2025 May-Jun.

NO ABSTRACT

PMID:40103835 | PMC:PMC11914739 | DOI:10.1016/j.xops.2025.100719

Front Med (Lausanne). 2025 Mar 4;12:1553970. doi: 10.3389/fmed.2025.1553970. eCollection 2025.

ABSTRACT

BACKGROUND: Critical illness medicine faces challenges such as high data complexity, large individual differences, and rapid changes in conditions. Artificial Intelligence (AI) technology, especially machine learning and deep learning, offers new possibilities for addressing these issues. By analyzing large amounts of patient data, AI can help identify diseases earlier, predict disease progression, and support clinical decision-making.

METHODS: In this study, scientific literature databases such as Web of Science were searched, and bibliometric methods along with visualization tools R-bibliometrix, VOSviewer 1.6.19, and CiteSpace 6.2.R4 were used to perform a visual analysis of the retrieved data.

RESULTS: This study analyzed 900 articles from 6,653 authors in 82 countries between 2005 and 2024. The United States is a major contributor in this field, with Harvard University having the highest betweenness centrality. Noseworthy PA is a core author in this field, and Frontiers in Cardiovascular Medicine and Diagnostics lead other journals in terms of the number of publications. Artificial Intelligence has tremendous potential in the identification and management of heart failure and sepsis.

CONCLUSION: The application of AI in critical illness holds great potential, particularly in enhancing diagnostic accuracy, personalized treatment, and clinical decision support. However, to achieve widespread application of AI technology in clinical practice, challenges such as data privacy, model interpretability, and ethical issues need to be addressed. Future research should focus on the transparency, interpretability, and clinical validation of AI models to ensure their effectiveness and safety in critical illness.

PMID:40103796 | PMC:PMC11914116 | DOI:10.3389/fmed.2025.1553970

Data Brief. 2025 Feb 12;59:111379. doi: 10.1016/j.dib.2025.111379. eCollection 2025 Apr.

ABSTRACT

The economic development of many countries largely depends on tea plantations that suffer from diseases adversely affecting their productivity and quality. This study presents a high-resolution dataset aimed at advancing precision agriculture for managing tea garden diseases. The size of the dataset is 3960 images and pixel dimension is (1024 × 1024) of the images were collected by using smartphones. This dataset contains detailed images of Tea Leaf Blight, Tea Red Leaf Spot and Tea Red Scab maladies inflicted on tea leaves as well as environmental statistics and plant health. The images were captured and stored in JPG format. The main aim of this dataset is to provide tool for detection and classification of different types of tea garden disease. Applying this dataset will enable the development of early detection systems, best-practice care regimens, and enhanced general garden upkeep. A range of images presenting the most prevalent diseases afflicting tea plants are paired with images of healthy leaves to provide a comprehensive overview of all the circumstances that can arise in a tea plantation. Therefore, it can be used to automate diseases tracking, targeted pesticide spraying, and even the making of smart farm tools with development of smart agricultural tools hence enhancing sustainability and efficiency in tea production. This dataset not only provides a strong foundation for applying precision techniques in tea cultivation in agriculture, but also can become an invaluable asset to scientists studying the issues of tea production.

PMID:40103762 | PMC:PMC11914274 | DOI:10.1016/j.dib.2025.111379

Data Brief. 2025 Feb 12;59:111387. doi: 10.1016/j.dib.2025.111387. eCollection 2025 Apr.

ABSTRACT

The rapid growth in wireless technology has revolutionized the way of living but at the same time, raising security concerns of unauthorized access of spectrum, both military and commercial sectors. The subject of Radio Frequency (RF) fingerprinting has got special attention in recent years. Researchers proposed various datasets of radio signals of different types of devices (drones, cell phones, IoT, and Radar). However, presently there is no freely available dataset on walkie-talkies/commercial radios. To fill out the void, we present an innovative dataset including more than 2700 radio signals captured from 27 radios located in an indoor multipath environment. This dataset can enhance the security of the communication channels by providing the possibility to analyse and detect any unauthorized source of transmission. Furthermore, we also propose two innovative deep learning models named Light Weight 1DCNN and Light Weight Bivariate 1DCNN, for efficient data processing and learning patterns from the complex dataset of radio signals.

PMID:40103755 | PMC:PMC11914181 | DOI:10.1016/j.dib.2025.111387