J Sci Food Agric. 2025 May 5. doi: 10.1002/jsfa.14331. Online ahead of print.

ABSTRACT

BACKGROUND: Automated fruit defect detection plays a critical role in improving postharvest quality assessment and supporting decision-making in agricultural supply chains. Guava defect detection presents specific challenges because of diverse disease types, varying maturity levels and inconsistent environmental conditions. Although existing you only look once (YOLO)-based models have shown promise in agricultural detection tasks, they often face limitations in balancing detection accuracy, inference speed and computational efficiency, particularly in resource-constrained settings. This study addresses this gap by evaluating four YOLO models (YOLOv8s, YOLOv5s, YOLOv9s and YOLOv11n) for detecting defective guava fruits across five diseases (scab, canker, chilling injury, mechanical damage and rot), three maturity levels (mature, half-mature and immature) and healthy fruits.

RESULTS: Diverse datasets facilitated robust training and evaluation. YOLOv11n achieved the highest mAP50-95 (98.0%) and exhibited bounding box loss (0.0565), classification loss (0.2787), inference time (3.9 milliseconds) and detection speed (255 FPS). YOLOv5s had the highest precision (94.9%), while YOLOv9s excelled in recall (96.2%). YOLOv8s offered a balanced performance across metrics. YOLOv11n outperformed all models with a lightweight architecture (2.6 million parameters) and low computational cost (6.3 giga floating-point operations per second), making it suitable for resource-constrained applications.

CONCLUSION: These results highlight YOLOv11n's potential for agricultural applications, such as automated defect detection and quality control, which require high accuracy and real-time performance across diverse conditions. This analysis provides insights into deploying YOLO models for agricultural quality assessment to enhance the efficiency and reliability of postharvest management. © 2025 Society of Chemical Industry.

PMID:40322977 | DOI:10.1002/jsfa.14331

Bioinformation. 2025 Feb 28;21(2):173-176. doi: 10.6026/973206300210173. eCollection 2025.

ABSTRACT

Artificial intelligence (AI) driven automated image analysis accurately tracks orthodontic tooth movement by reducing reliance on time-consuming manual assessments. AI achieved 92% precision with a 0.25 mm error margin and a strong correlation (r = 0.94, p < 0.001) to manual measurements in a study of 100 patients. AI analysis took 3 seconds per image set, significantly faster than the 7-minute manual process (p < 0.001). Orthodontists rated AI reliability at 4.7/5, with 86% preferring AI-assisted monitoring. Thus, AI enhances treatment efficiency, standardization, and clinical decision-making.

PMID:40322709 | PMC:PMC12044183 | DOI:10.6026/973206300210173

Bioinformation. 2025 Feb 28;21(2):105-109. doi: 10.6026/973206300210105. eCollection 2025.

ABSTRACT

The integration of Artificial Intelligence (AI) in to the field of medicine is offering a new-age of updated diagnostics, prediction and treatment across multiple fields, addressing systemic disease including viral infections and cancer. The fields of Oral Pathology, Dermatology, Psychiatry and Cardiology are shifting towards integrating these algorithms to improve health outcomes. AI trained on biomarkers (e.g. salivary cf DNA) has shown to uncover the genetic linkage to disease and symptom susceptibility. AI-enhanced imaging has increased sensitivity in cancer and lesion detection, as well as detecting functional abnormalities not clinically identified. The integration of AI across fields enables a systemic approach to understanding chronic inflammation, a central driver in conditions like cardiovascular disease, diabetes and neuropsychiatric disorders. We propose that through the use of imaging data with biomarkers like cytokines and genetic variants, AI models can better trace the effects of inflammation on immune and metabolic disruptions. This can be applied to the pandemic response, where AI can model the cascading effects of systemic dysfunctions, refine predictions of severe outcomes and guide targeted interventions to mitigate the multi-systemic impacts of pathogenic diseases.

PMID:40322698 | PMC:PMC12044186 | DOI:10.6026/973206300210105

Cureus. 2025 May 3;17(5):e83421. doi: 10.7759/cureus.83421. eCollection 2025 May.

ABSTRACT

Breast cancer remains one of the leading causes of mortality among women, particularly in low- and middle-income countries, where limited healthcare access and delayed diagnosis contribute to poor outcomes. Deep learning, especially convolutional neural networks (CNNs), has shown remarkable efficacy in breast cancer detection through automated image analysis, reducing reliance on manual interpretation. This study provides a comprehensive review of recent advancements in CNN-based breast cancer detection, evaluating deep learning architectures, feature extraction techniques, and optimization strategies. A comparative analysis of CNNs, recurrent neural networks (RNNs), and hybrid models highlights their strengths, limitations, and applicability in medical image classification. Using a dataset of 569 instances with 33 tumor morphology features, various deep learning architectures - including CNNs, long short-term memory networks (LSTMs), and multilayer perceptrons (MLPs) - were implemented, achieving classification accuracies between 89% and 98%. The study underscores the significance of data augmentation, transfer learning, and feature selection in improving model performance. Hybrid CNN-based models demonstrated superior predictive accuracy by capturing spatial and sequential dependencies within tumor feature sets. The findings support the potential of AI-driven breast cancer detection in clinical applications, reducing diagnostic errors and improving early detection rates. Future research should explore transformer-based models, federated learning, and explainable AI techniques to enhance interpretability, robustness, and generalization across diverse datasets.

PMID:40322605 | PMC:PMC12049196 | DOI:10.7759/cureus.83421

Clin Cosmet Investig Dermatol. 2025 Apr 29;18:1033-1044. doi: 10.2147/CCID.S508580. eCollection 2025.

ABSTRACT

BACKGROUND: Melasma is a prevalent pigmentary disorder characterized by treatment resistance and high recurrence. Existing assessment methods like the Melasma Area and Severity Index (MASI) are subjective and prone to inter-observer variability.

OBJECTIVE: This study aimed to develop an AI-assisted, real-time melasma severity classification framework based on deep learning and clinical facial images.

METHODS: A total of 1368 anonymized facial images were collected from clinically diagnosed melasma patients. After image preprocessing and MASI-based labeling, six CNN architectures were trained and evaluated using PyTorch. Model performance was assessed through accuracy, precision, recall, F1-score, AUC, and interpretability via Layer-wise Relevance Propagation (LRP).

RESULTS: GoogLeNet achieved the best performance, with an accuracy of 0.755 and an F1-score of 0.756. AUC values across severity levels reached 0.93 (mild), 0.86 (moderate), and 0.94 (severe). LRP analysis confirmed GoogLeNet's superior feature attribution.

CONCLUSION: This study presents a robust, interpretable deep learning model for melasma severity classification, offering enhanced diagnostic consistency. Future work will integrate multimodal data for more comprehensive assessment.

PMID:40322508 | PMC:PMC12049110 | DOI:10.2147/CCID.S508580

Biomed Opt Express. 2025 Apr 1;16(4):1732-1741. doi: 10.1364/BOE.557748. eCollection 2025 Apr 1.

ABSTRACT

This study evaluates the role of differential artery-vein (AV) analysis in optical coherence tomography angiography (OCTA) for treatment outcome prediction of diabetic macular edema (DME). Deep learning AV segmentation in OCTA enabled the robust extraction of quantitative AV features, including perfusion intensity density (PID), blood vessel density (BVD), vessel skeleton density (VSD), vessel area flux (VAF), blood vessel caliber (BVC), blood vessel tortuosity (BVT), and vessel perimeter index (VPI). Support vector machine (SVM) classifiers were employed to predict changes in best-corrected visual acuity (BCVA) and central retinal thickness (CRT). Comparative analysis revealed that differential AV analysis significantly enhanced prediction performance, with BCVA accuracy improved from 70.45% to 86.36% and CRT accuracy enhanced from 68.18% to 79.55% compared to traditional OCTA analysis. These findings underscore the potential of AV analysis as a transformative tool for advancing personalized therapeutic strategies and improving clinical decision-making in managing DME.

PMID:40322014 | PMC:PMC12047724 | DOI:10.1364/BOE.557748

Biomed Opt Express. 2025 Mar 26;16(4):1582-1601. doi: 10.1364/BOE.549880. eCollection 2025 Apr 1.

ABSTRACT

Diffuse optical tomography (DOT) performed using deep-learning allows high-speed reconstruction of tissue optical properties and could thereby enable image-guided scanning, e.g., to enhance clinical breast imaging. Previously published models are geometry-specific and, therefore, require extensive data generation and training for each use case, restricting the scanning protocol at the point of use. A transformer-based architecture is proposed to overcome these obstacles that encode spatially unstructured DOT measurements, enabling a single trained model to handle arbitrary scanning pathways and measurement density. The model is demonstrated with breast tissue-emulating simulated and phantom data, yielding - for 24 mm-deep absorptions (μ a ) and reduced scattering (μ s ') images, respectively - average RMSEs of 0.0095±0.0023 cm-1 and 1.95±0.78 cm-1, Sørensen-Dice coefficients of 0.55±0.12 and 0.67±0.1, and anomaly contrast of 79±10% and 93.3±4.6% of the ground-truth contrast, with an effective imaging speed of 14 Hz. The average absolute μ a and μ s ' values of homogeneous simulated examples were within 10% of the true values.

PMID:40322000 | PMC:PMC12047716 | DOI:10.1364/BOE.549880

Biomed Opt Express. 2025 Mar 21;16(4):1528-1545. doi: 10.1364/BOE.558675. eCollection 2025 Apr 1.

ABSTRACT

Recent advancements in cellular-resolution optical coherence tomography (OCT) have opened up possibilities for high-resolution and non-invasive clinical diagnosis. This study uses deep learning-based models on cross-sectional OCT images for in vivo human skin layers and keratinocyte nuclei segmentation. With U-Net as the basic framework, a 5-class segmentation model is developed. With deeply supervised learning objective functions, the global (skin layers) and local (nuclei) features were separately considered in designing our multi-class segmentation model to achieve an > 85% Dice coefficient accuracy through 5-fold cross-validation, enabling quantitative measurements for the healthy human skin structure. Specifically, we calculate the thickness of the stratum corneum, epidermis, and the cross-sectional area of keratinocyte nuclei as 22.71 ± 17.20 µm, 66.44 ± 11.61 µm, and 17.21 ± 9.33 µm2, respectively. These measurements align with clinical findings on human skin structures and can serve as standardized metrics for clinical assessment using OCT imaging. Moreover, we enhance the segmentation accuracy by addressing the limitations of microscopic system resolution and the variability in human annotations.

PMID:40321995 | PMC:PMC12047727 | DOI:10.1364/BOE.558675

ArXiv [Preprint]. 2025 Apr 16:arXiv:2504.12480v1.

ABSTRACT

Reservoir computers (RCs) provide a computationally efficient alternative to deep learning while also offering a framework for incorporating brain-inspired computational principles. By using an internal neural network with random, fixed connections$-$the 'reservoir'$-$and training only the output weights, RCs simplify the training process but remain sensitive to the choice of hyperparameters that govern activation functions and network architecture. Moreover, typical RC implementations overlook a critical aspect of neuronal dynamics: the balance between excitatory and inhibitory (E-I) signals, which is essential for robust brain function. We show that RCs characteristically perform best in balanced or slightly over-inhibited regimes, outperforming excitation-dominated ones. To reduce the need for precise hyperparameter tuning, we introduce a self-adapting mechanism that locally adjusts E/I balance to achieve target neuronal firing rates, improving performance by up to 130% in tasks like memory capacity and time series prediction compared with globally tuned RCs. Incorporating brain-inspired heterogeneity in target neuronal firing rates further reduces the need for fine-tuning hyperparameters and enables RCs to excel across linear and non-linear tasks. These results support a shift from static optimization to dynamic adaptation in reservoir design, demonstrating how brain-inspired mechanisms improve RC performance and robustness while deepening our understanding of neural computation.

PMID:40321946 | PMC:PMC12047930

Heart Rhythm O2. 2025 Jan 21;6(4):473-480. doi: 10.1016/j.hroo.2025.01.008. eCollection 2025 Apr.

ABSTRACT

BACKGROUND: Rotors and focal ectopies, or "drivers," are hypothesized mechanisms of persistent atrial fibrillation (AF). Machine learning algorithms have been used to identify these drivers, but the limited size of current driver data sets constrains their performance.

OBJECTIVE: We proposed that pretraining using unsupervised learning on a substantial data set of unlabeled electrograms could enhance classifier accuracy when applied to a smaller driver data set.

METHODS: We used a SimCLR-based framework to pretrain a residual neural network on 113,000 unlabeled 64-electrode measurements from a canine model of AF. The network was then fine-tuned to identify drivers from intracardiac electrograms. Various augmentations, including cropping, Gaussian blurring, and rotation, were applied during pretraining to improve the robustness of the learned representations.

RESULTS: Pretraining significantly improved driver detection accuracy compared with a non-pretrained network (80.8% vs 62.5%). The pretrained network also demonstrated greater resilience to reductions in training data set size, maintaining higher accuracy even with a 30% reduction in data. Gradient-weighted Class Activation Mapping analysis revealed that the network's attention aligned well with manually annotated driver regions, suggesting that the network learned meaningful features for driver detection.

CONCLUSION: This study demonstrates that contrastive pretraining can enhance the accuracy of driver detection algorithms in AF. The findings support the broader application of transfer learning to other electrogram-based tasks, potentially improving outcomes in clinical electrophysiology.

PMID:40321744 | PMC:PMC12047512 | DOI:10.1016/j.hroo.2025.01.008

Pol J Radiol. 2025 Mar 14;90:e124-e137. doi: 10.5114/pjr/200628. eCollection 2025.

ABSTRACT

PURPOSE: Tuberculosis (TB) continues to be a major cause of death from infectious diseases globally. TB is treatable with antibiotics, but it is often misdiagnosed or left untreated, particularly in rural and resource-constrained regions. While chest X-rays are a key tool in TB diagnosis, their effectiveness is hindered by the variability in radiological presentations and the lack of trained radiologists in high-prevalence areas. Deep learning-based imaging techniques offer a promising approach to computer-aided diagnosis for TB, enabling precise and timely detection while alleviating the burden on healthcare professionals. This study aims to enhance TB detection in chest X-ray images by developing deep learning models. We have observed upper and lower lobe consolidation, pleural effusion, calcification, cavity formation and military nodules. A proposed preprocessing technique has been also introduced in our work based on gamma correction and gradient based technique for contrast enhancement. We leverage the Res-UNet architecture for image segmentation and introduce a novel deep learning network for classification, targeting improved accuracy and precision in diagnostic performance.

MATERIAL AND METHODS: A Res-UNet segmentation model was trained using 704 chest X-ray images sourced from the Montgomery County and Shenzhen Hospital datasets. Following training, the model was applied to segment lung regions in 1400 chest X-ray scans, encompassing both TB cases and normal controls, obtained from the National Institute of Allergy and Infectious Diseases (NIAID) TB Portal program dataset. The segmented lung regions were subsequently classified as either TB or normal using a deep learning model. A gradient based technique was used for contrast enhancement by capturing intensity changes in image by comparing each pixel with its neighbour with pyramid reduction unique mapping and histogram matching along with gamma correction is used. This integrated approach of segmentation and classification aims to enhance the accuracy and precision of TB detection in chest X-ray images. Classification of segmented images was done using customised convolutional neural network, and visualisation was done using Grad-CAM.

RESULTS: The Res-UNet model demonstrated excellent performance for segmentation, achieving an accuracy of 98.18%, recall of 98.40%, precision of 97.45%, F1-score of 97.97%, Dice coefficient of 96.33%, and Jaccard index of 96.05%. Similarly, the classification model exhibited outstanding results, with a classification accuracy of 99.45%, precision of 99.29%, recall of 99.29%, F1-score of 99.29%, and an AUC of 99.9%. Enhanced gradient based method showed ambe of 16.51, entropy of 6.7370, CII of 86.80, psnr of 28.71, ssim of 86.83 which are quite satisfactory.

CONCLUSIONS: The findings demonstrate the efficiency of our system in diagnosing TB from chest X-rays, potentially surpassing clinician-level precision. This underscores its effectiveness as a diagnostic tool, particularly in resourcelimited settings with restricted access to radiological expertise. Additionally, the modified Res-UNet model demonstrated superior performance compared to the standard U-Net, highlighting its potential for achieving greater diagnostic accuracy.

PMID:40321710 | PMC:PMC12049158 | DOI:10.5114/pjr/200628

Water Res. 2025 Apr 22;282:123711. doi: 10.1016/j.watres.2025.123711. Online ahead of print.

ABSTRACT

Urban flood control and drainage systems often face significant challenges in coordinating municipal drainage with river-lake flood prevention during flood seasons. Rising river levels can create backwater effects, which substantially increase urban flood risks. Traditional water management approaches are limited by delayed monitoring data updates, slow flood forecasting processes, and inadequate decision support, making it difficult to address the complex, multi-objective demands of flood control. These limitations exacerbate flooding threats and hamper effective urban flood management. To address these challenges, a digital twin experimental platform for river and lake water systems was developed to enhance the comprehensive management of urban flood control and drainage. The platform integrates an engineering entity, a backend system, and a digital twin component. Real-time data acquisition and virtual-real interactions between physical facilities and the digital twin were achieved using Programmable Logic Controller (PLC) technology, while the Unity3D engine enabled advanced visualization and data rendering. Furthermore, a novel model incorporating deep learning and a multi-objective optimization algorithm was proposed to optimize drainage pump scheduling rules. A comparative analysis was conducted to evaluate flood risks and operation and maintenance costs before and after optimization. The results demonstrated that the platform was well-designed for comprehensive flood protection and drainage management. The NSE coefficients for river and lake water levels exceeded 95.18 %, and the relative error in pump operation times remained below 4.11 % across various scenarios involving river inflows and drainage operations. The backwater effect at drainage outlets was primarily driven by river flow and downstream lake levels. The optimization strategy effectively balanced water level control and operational objectives, reducing water level targets by 24.99 %, 40.36 %, and 51.61 % under different scenarios. This framework not only offers innovative solutions for urban flood management but also provides strong technical support for optimizing flood control and drainage system operations.

PMID:40319783 | DOI:10.1016/j.watres.2025.123711

Comput Biol Med. 2025 May 3;192(Pt A):110279. doi: 10.1016/j.compbiomed.2025.110279. Online ahead of print.

ABSTRACT

BACKGROUND: Oral health plays a vital role in our daily lives, affecting essential activities like eating, speaking, and smiling. Poor oral health can lead to significant social, psychological, and physical consequences, which makes early and accurate diagnosis incredibly important. Recent advances in artificial intelligence (AI) are opening new doors in oral health care, offering faster, more accurate ways to identify dental issues and improve overall care.

METHODS: This paper uses RGB color photography to introduce a non-linear Fuzzy Rank-based Ensemble Deep Learning model (NFR-EDL) for diagnosing oral and dental diseases. The model utilizes four deep Convolutional Neural Network (CNN) base models to analyze high-resolution color images of the oral cavity. The CNN base models are initially trained to generate confidence scores, which are subsequently mapped onto distinct functions with varying concavities, resulting in non-linear fuzzy ranks. Then, these ranks are combined into a final score to minimize the deviation from expected results. This method aims to provide accurate, reliable identification of oral and dental disease diagnosis by fusing many base models and considering uncertainty in decision-making while utilizing the rich visual information available in RGB images.

RESULTS: The experimental results demonstrate that the proposed NFR-EDL model achieves accuracies of 97.08 %, 84.00 %, 89.86 %, and 94.66 % on the Kaggle, MOD, ODSI-DB, and OaDD datasets, respectively. These results demonstrate the model's exceptional accuracy and effectiveness in diagnosing oral and dental diseases, outperforming existing techniques and enhancing diagnostic reliability in clinical settings.

CONCLUSION: Deploying the NFR-EDL model in clinical settings offers a highly accurate and reliable tool for diagnosing oral and dental diseases, enhancing early detection, personalizing patient care, and reducing diagnostic errors to ultimately improve patient outcomes and the efficiency of dental care delivery. This approach reduces uncertainty in decision-making, ensuring that diagnoses are made with high confidence.

PMID:40319757 | DOI:10.1016/j.compbiomed.2025.110279

J Cheminform. 2025 May 4;17(1):67. doi: 10.1186/s13321-025-01019-y.

ABSTRACT

Post-translational modifications (PTMs) play a crucial role in allowing cells to expand the functionality of their proteins and adaptively regulate their signaling pathways. Defects in PTMs have been linked to numerous developmental disorders and human diseases, including cancer, diabetes, heart, neurodegenerative and metabolic diseases. PTMs are important targets in drug discovery, as they can significantly influence various aspects of drug interactions including binding affinity. The structural consequences of PTMs, such as phosphorylation-induced conformational changes or their effects on ligand binding affinity, have historically been challenging to study on a large scale, primarily due to reliance on experimental methods. Recent advancements in computational power and artificial intelligence, particularly in deep learning algorithms and protein structure prediction tools like AlphaFold3, have opened new possibilities for exploring the structural context of interactions between PTMs and drugs. These AI-driven methods enable accurate modeling of protein structures including prediction of PTM-modified regions and simulation of ligand-binding dynamics on a large scale. In this work, we identified small molecule binding-associated PTMs that can influence drug binding across all human proteins listed as small molecule targets in the DrugDomain database, which we developed recently. 6,131 identified PTMs were mapped to structural domains from Evolutionary Classification of Protein Domains (ECOD) database.Scientific contribution: Using recent AI-based approaches for protein structure prediction (AlphaFold3, RoseTTAFold All-Atom, Chai-1), we generated 14,178 models of PTM-modified human proteins with docked ligands. Our results demonstrate that these methods can predict PTM effects on small molecule binding, but precise evaluation of their accuracy requires a much larger benchmarking set. We also found that phosphorylation of NADPH-Cytochrome P450 Reductase, observed in cervical and lung cancer, causes significant structural disruption in the binding pocket, potentially impairing protein function. All data and generated models are available from DrugDomain database v1.1 ( http://prodata.swmed.edu/DrugDomain/ ) and GitHub ( https://github.com/kirmedvedev/DrugDomain ). This resource is the first to our knowledge in offering structural context for small molecule binding-associated PTMs on a large scale.

PMID:40320551 | DOI:10.1186/s13321-025-01019-y

Sci Rep. 2025 May 4;15(1):15602. doi: 10.1038/s41598-025-97656-2.

ABSTRACT

Mobile robots are used in various fields, from deliveries to search and rescue applications. Different types of sensors are mounted on the robot to provide accurate navigation and, thus, allow successful completion of its task. In real-world scenarios, due to environmental constraints, the robot frequently relies only on its inertial sensors. Therefore, due to noises and other error terms associated with the inertial readings, the navigation solution drifts in time. To mitigate the inertial solution drift, we propose the MoRPINet framework consisting of a neural network to regress the robot's travelled distance. To this end, we require the mobile robot to maneuver in a snake-like slithering motion to encourage nonlinear behavior. MoRPINet was evaluated using a dataset of 290 minutes of inertial recordings during field experiments and showed an improvement of 33% in the positioning error over other state-of-the-art methods for pure inertial navigation.

PMID:40320468 | DOI:10.1038/s41598-025-97656-2

Sci Rep. 2025 May 5;15(1):15616. doi: 10.1038/s41598-025-91518-7.

ABSTRACT

Rockburst is a significant safety threat in coal mining, influenced by complex nonlinear dynamic characteristics and multi-factor coupling. This study proposes a rockburst risk prediction method based on the SSA-CNN-MoLSTM-Attention model. The model integrates the local feature extraction capability of convolutional neural networks (CNN), the temporal modeling advantages of the modified long short-term memory network (MoLSTM), and the enhanced feature recognition capability of the attention mechanism. Additionally, the sparrow search algorithm (SSA) is employed to optimize hyperparameters, further improving the model's performance. Unlike traditional approaches that rely on time-axis-based analysis, this study uses the working face advancement distance as the basis for prediction, which better reveals the potential spatial correlations of rockburst occurrences, aligning with engineering practice needs.Validation using microseismic monitoring data from a coal mine demonstrates that the proposed model achieves a prediction accuracy of 93.62% and an F1-score of 93.54%. The model outperforms traditional methods in mean absolute error (MAE) and root mean square error (RMSE), providing effective insights and a reference for rockburst risk assessment and disaster prevention in mining operations.

PMID:40320457 | DOI:10.1038/s41598-025-91518-7

Sci Rep. 2025 May 4;15(1):15594. doi: 10.1038/s41598-025-00546-w.

ABSTRACT

This study aims to conduct a comprehensive and in-depth analysis of the marketing performance of e-commerce live broadcast platforms based on big data management technology and deep learning. Firstly, by synthesizing large-scale datasets and surveys, the study constructs a series of performance evaluation indicators including user participation, content quality, commodity sales effect, user satisfaction, and platform promotion effect. Secondly, the weight of each indicator is finally determined through the indicator screening of the expert scoring method. Finally, the experimental design and implementation steps such as data collection, experimental environment setting, parameter setting, and performance evaluation are introduced in detail. Through the training and evaluation of the Back Propagation Neural Network (BPNN), each secondary indicator's adjusted weight value and global ranking are obtained, providing a scientific basis for subsequent management opinions. The research results emphasize the importance of comments and ratings, purchase conversion rate, advertising click-through rate, and other indicators in improving user satisfaction, promoting sales, and effective promotion. Overall, this study provides a clear direction for an e-commerce live broadcast platform to optimize user experience, improve sales performance, and strengthen brand promotion.

PMID:40320449 | DOI:10.1038/s41598-025-00546-w

Sci Rep. 2025 May 4;15(1):15614. doi: 10.1038/s41598-025-00516-2.

ABSTRACT

Lung cancer has been stated as one of the prevalent killers of cancer up to this present time and this clearly underlines the rationale for early diagnosis to enhance life expectancy of patients afflicted with the condition. The reasons behind the usage of the transformer and deep learning classifiers for the detection of lung cancer include accuracy, robustness along with the capability to handle and evaluate large data sets and much more. Such models can be more complex and can help to utilize multiple modalities of data to give extensive information that will be critical in ascertaining the right diagnosis at the right time. However, the existing works encounter several limitations including reliance on large annotated data, overfitting, high computation complexity, and interpretability. Third, the issue of the stability of these models' performance when applied to actual clinical datasets is still an open question; this is an even bigger issue that will greatly reduce the actual utilization of these models in clinical practice. To tackle these, we develop a novel Cancer Nexus Synergy (CanNS), which applies of A. Swin-Transformer UNet (SwiNet) Model for segmentation, Xception-LSTM GAN (XLG) CancerNet for classification, and Devilish Levy Optimization (DevLO) for fine-tuning parameters. This paper breaks new ground in that the presented elements are incorporated in a manner that co-operatively elevates the diagnostic capabilities while at the same time being computationally light and resilient. These are SwiNet for segmented analysis, XLG CancerNet for precise classification of the cases, and DevLO that optimizes the parameters of the lung cancer detection system, making the system more sensible and efficient. The performance outcomes indicate that the CanNS framework enhances the detection's accuracy, sensitivity, and specificity compared to the previous approaches.

PMID:40320438 | DOI:10.1038/s41598-025-00516-2

Sci Rep. 2025 May 4;15(1):15608. doi: 10.1038/s41598-025-00563-9.

ABSTRACT

White blood cell (WBC) classification is a crucial step in assessing a patient's health and validating medical treatment in the medical domain. Hence, efficient computer vision solutions to the classification of WBC will be an effective aid to medical practitioners. Computer-aided diagnosis (CAD) reduces manual intervention, avoids errors, speeds up medical analysis, and provides accurate medical reports. Though a lot of research has been taken up to develop deep learning models for efficient classification of WBCs, there is still scope for improvement to support the data insufficiency issue in medical data sets. Data augmentation and normalization techniques increase the quantity of data but don't enhance the quality of the data. Hence, deep learning models though performing well can still be made efficient and effective when quality data is fused along with the available image dataset. This paper aims to utilize domain knowledge and image data to improve the classification performance of pre-trained models namely Inception V3, DenseNet 121, ResNet 50, MobileNet V2, and VGG 16. The models performance, with and without domain knowledge infused, is analyzed on the BCCD and LISC datasets. On the BCCD dataset, the average accuracies increased from 82.7%, 98.8%, 98.38%, 98.56%, and 98.5%-99.38%, 99.05%, 99.05%, 98.67%, and 98.75% for Inception V3, DenseNet 121, ResNet 50, MobileNet V2, and VGG 16, respectively. Similarly, on the LISC dataset, the accuracies improved from 86.76%, 92.2%, 91.76%, 92.8%, and 94.4%-92.05%, 95.88%, 95.58%, 95.2%, and 95.2%, respectively.

PMID:40320432 | DOI:10.1038/s41598-025-00563-9

Network. 2025 May 4:1-35. doi: 10.1080/0954898X.2025.2500046. Online ahead of print.

ABSTRACT

Brain tumour classification plays a significant role in improving patient care, treatment planning, and enhancing the overall healthcare system's effectiveness. This article presents a ResNet framework optimized using Henry gas solubility optimization (HGSO) for the classification of brain tumours, resulting in improved classification performance in magnetic resonance images (MRI). Two variants of the deep residual neural network, namely ResNet-18 and ResNet-50, are trained on the MRI training dataset. The four critical hyperparameters of the ResNet model: momentum, initial learning rate, maximum epochs, and validation frequency are tuned to obtain optimal values using HGSO algorithm. Subsequently, the optimized ResNet model is evaluated using two separate databases: Database1, comprising four tumour classes, and Database2, with three tumour classes. The performance is assessed using accuracy, sensitivity, specificity, precision, and F-score. The highest classification accuracy of 0.9825 is attained using the proposed optimized ResNet-50 framework on Database1. Moreover, the Gradient-weighted Class Activation Mapping (GRAD-CAM) algorithm is utilized to enhance the understanding of deep neural networks by highlighting the regions that are influential in making a particular classification decision. Grad-CAM heatmaps confirm the model focuses on relevant tumour features, not image artefacts. This research enhances MRI brain tumour classification via deep learning optimization strategies.

PMID:40320295 | DOI:10.1080/0954898X.2025.2500046