Bone Joint J. 2025 Jun 1;107-B(6):582-586. doi: 10.1302/0301-620X.107B6.BJJ-2024-1567.R1.

ABSTRACT

The deployment of AI in medical imaging, particularly in areas such as fracture detection, represents a transformative advancement in orthopaedic care. AI-driven systems, leveraging deep-learning algorithms, promise to enhance diagnostic accuracy, reduce variability, and streamline workflows by analyzing radiograph images swiftly and accurately. Despite these potential benefits, the integration of AI into clinical settings faces substantial barriers, including slow adoption across health systems, technical challenges, and a major lag between technology development and clinical implementation. This commentary explores the role of AI in healthcare, highlighting its potential to enhance patient outcomes through more accurate and timely diagnoses. It addresses the necessity of bridging the gap between AI innovation and practical application. It also emphasizes the importance of implementation science in effectively integrating AI technologies into healthcare systems, using frameworks such as the Consolidated Framework for Implementation Research and the Knowledge-to-Action Cycle to guide this process. We call for a structured approach to address the challenges of deploying AI in clinical settings, ensuring that AI's benefits translate into improved healthcare delivery and patient care.

PMID:40449898 | DOI:10.1302/0301-620X.107B6.BJJ-2024-1567.R1

Bone. 2025 May 29:117550. doi: 10.1016/j.bone.2025.117550. Online ahead of print.

ABSTRACT

OBJECTIVE: Despite the established association between chronic obstructive pulmonary disease (COPD) severity and risk of osteoporosis, even after accounting for the known shared confounding variables (e.g., age, smoking, history of exacerbations, steroid use), there is paucity of data on bone loss among mild to moderate COPD, which is more prevalent in the general population.

METHODS: We conducted a longitudinal analysis using data from the Multi-Ethnic Study of Atherosclerosis. Participants with chest CT at Exam 5 (2010-2012) and Exam 6 (2016-2018) were included. Mild to moderate COPD was defined as forced expiratory volume in 1 s (FEV1) to forced vital capacity ratio of <0.70 and FEV1 of 50 % or higher. Vitamin D deficiency was defined as serum vitamin D < 20 ng/mL. We utilized a validated deep learning algorithm to perform automated multilevel segmentation of vertebral bodies (T1-T10) from chest CT and derive 3D volumetric thoracic vertebral BMD measurements at Exam 5 and 6.

RESULTS: Of the 1226 participants, 173 had known mild to moderate COPD at baseline, while 1053 had no known COPD. After adjusting for age, race/ethnicity, sex, body mass, index, bisphosphonate use, alcohol consumption, smoking, diabetes, physical activity, C-reactive protein and vitamin D deficiency, mild to moderate COPD was associated with faster decline in BMD (estimated difference, β = -0.38 g/cm3/year; 95 % CI: -0.74, -0.02). A significant interaction between COPD and vitamin D deficiency (p = 0.001) prompted stratified analyses. Among participants with vitamin D deficiency (47 % of participants), COPD was associated with faster decline in BMD (-0.64 g/cm3/year; 95 % CI: -1.17 to -0.12), whereas no significant association was observed among those with normal vitamin D in both crude and adjusted models.

CONCLUSIONS: Mild to moderate COPD is associated with longitudinal declines in vertebral BMD exclusively in participants with vitamin D deficiency over 6-year follow-up. Vitamin D deficiency may play a crucial role in bone loss among patients with mild to moderate COPD.

PMID:40449861 | DOI:10.1016/j.bone.2025.117550

Plant Methods. 2025 May 30;21(1):75. doi: 10.1186/s13007-025-01393-6.

ABSTRACT

Wheat, a major global cereal crop, is prone to quality degradation from early sprouting when stored improperly, resulting in substantial economic losses. Traditional methods for detecting early sprouting are labor-intensive and destructive, underscoring the need for rapid, non-destructive alternatives. Terahertz (THz) technology provides a promising solution due to its ability to perform non-invasive imaging of internal structures. However, current THz imaging techniques are limited by low image resolution, which restricts their practical application. We address these challenges by proposing an advanced deep learning framework for THz image classification of early sprouting wheat. We first develop an Enhanced Super-Resolution Generative Adversarial Network (AESRGAN) to improve the resolution of THz images, integrating an attention mechanism to focus on critical image regions. This model achieves a 0.76 dB improvement in Peak Signal-to-Noise Ratio (PSNR). Subsequently, we introduce the EfficientViT-based YOLO V8 classification model, incorporating a Depthwise Separable Attention (C2F-DSA) module, and further optimize the model using the Gazelle Optimization Algorithm (GOA). Experimental results demonstrate the GOA-EViTDSA-YOLO model achieves an accuracy of 97.5% and a mean Average Precision (mAP) of 0.962. The model is efficient and significantly enhances the classification of early sprouting wheat compared to other deep learning models.

PMID:40448208 | DOI:10.1186/s13007-025-01393-6

BMC Med Imaging. 2025 May 30;25(1):200. doi: 10.1186/s12880-025-01746-6.

ABSTRACT

PURPOSE: To compare the image quality and pulmonary nodule detectability and measurement accuracy between deep learning reconstruction (DLR) and hybrid iterative reconstruction (HIR) of chest ultra-low-dose CT (ULDCT).

MATERIALS AND METHODS: Participants who underwent chest standard-dose CT (SDCT) followed by ULDCT from October 2020 to January 2022 were prospectively included. ULDCT images reconstructed with HIR and DLR were compared with SDCT images to evaluate image quality, nodule detection rate, and measurement accuracy using a commercially available deep learning-based nodule evaluation system. Wilcoxon signed-rank test was used to evaluate the percentage errors of nodule size and nodule volume between HIR and DLR images.

RESULTS: Eighty-four participants (54 ± 13 years; 26 men) were finally enrolled. The effective radiation doses of ULDCT and SDCT were 0.16 ± 0.02 mSv and 1.77 ± 0.67 mSv, respectively (P < 0.001). The mean ± standard deviation of the lung tissue noises was 61.4 ± 3.0 HU for SDCT, 61.5 ± 2.8 HU and 55.1 ± 3.4 HU for ULDCT reconstructed with HIR-Strong setting (HIR-Str) and DLR-Strong setting (DLR-Str), respectively (P < 0.001). A total of 535 nodules were detected. The nodule detection rates of ULDCT HIR-Str and ULDCT DLR-Str were 74.0% and 83.4%, respectively (P < 0.001). The absolute percentage error in nodule volume from that of SDCT was 19.5% in ULDCT HIR-Str versus 17.9% in ULDCT DLR-Str (P < 0.001).

CONCLUSION: Compared with HIR, DLR reduced image noise, increased nodule detection rate, and improved measurement accuracy of nodule volume at chest ULDCT.

CLINICAL TRIAL NUMBER: Not applicable.

PMID:40448068 | DOI:10.1186/s12880-025-01746-6

Sci Rep. 2025 May 30;15(1):19103. doi: 10.1038/s41598-025-04061-w.

ABSTRACT

The Internet of Vehicles (IoV) is crucial in facilitating secure and efficient vehicle-infrastructure communication. Nevertheless, with an increasing reliance on the IoV in modern logistics and intelligent fleet systems, cyberattacks on vital supply chain information pose a far greater threat. This research presents the ASCON, a low-power cryptographic algorithm, with the Message Queued Telemetry Transport (MQTT) protocol for secure IoV communications. Integration of a deep learning model that is suited for real-time anomaly detection and breach prediction. The novelty of this study is the hybrid framework that uses lightweight cryptographic methods coupled with deep learning-based threat protection. Therefore, it is resilient against a wide range of cyber-attacks, including password cracking, authentication compromises, brute-force attacks, differential cryptanalysis, and Zig-Zag attacks. The system employs Raspberry Pi boards with authentic industrial vehicluar dataset and offers a remarkable encryption rate of 0.025 s, takes 0.003 s for hash generation, and detection of tampering takes 0.002 s. By bridging the gap between high-level cryptography and proactive and smart security analytics, this work not only fortifies fleet management systems but also makes substantial contributions to the overall objectives of enhancing safety, sustainability, and operational robustness in autonomous vehicle networks.

PMID:40447743 | DOI:10.1038/s41598-025-04061-w

Commun Biol. 2025 May 30;8(1):836. doi: 10.1038/s42003-025-08184-8.

ABSTRACT

Multiplexed imaging techniques require identifying different cell types in the tissue. To utilize their potential for cellular and molecular analysis, high throughput and accurate analytical approaches are needed in parsing vast amounts of data, particularly in clinical settings. Nuclear segmentation errors propagate in all downstream steps of cell phenotyping and single-cell spatial analyses. Here, we benchmark and compare the nuclear segmentation tools commonly used in multiplexed immunofluorescence data by evaluating their performance across 7 tissue types encompassing ~20,000 labeled nuclei from human tissue samples. Pre-trained deep learning models outperform classical nuclear segmentation algorithms. Overall, Mesmer is recommended as it exhibits the highest nuclear segmentation accuracy with 0.67 F1-score at an IoU threshold of 0.5 on our composite dataset. Pre-trained StarDist model is recommended in case of limited computational resources, providing ~12x run time improvement with CPU compute and ~4x improvement with the GPU compute over Mesmer, but it struggles in dense nuclear regions.

PMID:40447729 | DOI:10.1038/s42003-025-08184-8

BMC Med Inform Decis Mak. 2025 May 30;25(1):200. doi: 10.1186/s12911-025-03029-0.

ABSTRACT

PURPOSE: This study aims to develop a deep learning framework that leverages modality imputation and subregion segmentation to improve grading accuracy in high-grade gliomas.

MATERIALS AND METHODS: A retrospective analysis was conducted using data from 1,251 patients in the BraTS2021 dataset as the main cohort and 181 clinical cases collected from a medical center between April 2013 and June 2018 (51 years ± 17; 104 males) as the external test set. We propose a PatchGAN-based modality imputation network with an Aggregated Residual Transformer (ART) module combining Transformer self-attention and CNN feature extraction via residual links, paired with a U-Net variant for segmentation. Generative accuracy used PSNR and SSIM for modality conversions, while segmentation performance was measured with DSC and HD95 across necrotic core (NCR), edema (ED), and enhancing tumor (ET) regions. Senior radiologists conducted a comprehensive Likert-based assessment, with diagnostic accuracy evaluated by AUC. Statistical analysis was performed using the Wilcoxon signed-rank test and the DeLong test.

RESULTS: The best source-target modality pairs for imputation were T1 to T1ce and T1ce to T2 (p < 0.001). In subregion segmentation, the overall DSC was 0.878 and HD95 was 19.491, with the ET region showing the highest segmentation accuracy (DSC: 0.877, HD95: 12.149). Clinical validation revealed an improvement in grading accuracy by the senior radiologist, with the AUC increasing from 0.718 to 0.913 (P < 0.001) when using the combined imputation and segmentation models.

CONCLUSION: The proposed deep learning framework improves high-grade glioma grading by modality imputation and segmentation, aiding the senior radiologist and offering potential to advance clinical decision-making.

PMID:40448035 | DOI:10.1186/s12911-025-03029-0

BMC Med Imaging. 2025 May 30;25(1):197. doi: 10.1186/s12880-025-01737-7.

ABSTRACT

OBJECTIVE: This study aimed to construct a novel model, Multi-Spatial Attention U-Net (MSAU-Net) by incorporating our proposed Multi-Spatial Attention (MSA) block into the U-Net for the automated segmentation of the gallbladder on CT images.

METHODS: The gallbladder dataset consists of CT images of retrospectively-collected 152 liver cancer patients and corresponding ground truth delineated by experienced physicians. Our proposed MSAU-Net model was transformed into two versions V1(with one Multi-Scale Feature Extraction and Fusion (MSFEF) module in each MSA block) and V2 (with two parallel MSEFE modules in each MSA blcok). The performances of V1 and V2 were evaluated and compared with four other derivatives of U-Net or state-of-the-art models quantitatively using seven commonly-used metrics, and qualitatively by comparison against experienced physicians' assessment.

RESULTS: MSAU-Net V1 and V2 models both outperformed the comparative models across most quantitative metrics with better segmentation accuracy and boundary delineation. The optimal number of MSA was three for V1 and two for V2. Qualitative evaluations confirmed that they produced results closer to physicians' annotations. External validation revealed that MSAU-Net V2 exhibited better generalization capability.

CONCLUSION: The MSAU-Net V1 and V2 both exhibited outstanding performance in gallbladder segmentation, demonstrating strong potential for clinical application. The MSA block enhances spatial information capture, improving the model's ability to segment small and complex structures with greater precision. These advantages position the MSAU-Net V1 and V2 as valuable tools for broader clinical adoption.

PMID:40448013 | DOI:10.1186/s12880-025-01737-7

Discov Oncol. 2025 May 30;16(1):966. doi: 10.1007/s12672-025-02771-1.

ABSTRACT

Cancer has increasingly been recognized as a genetic disease, influenced by lifestyle changes, dietary patterns, and environmental pollutants. Lung cancer remains one of the most lethal malignancies worldwide, necessitating precise diagnostic and therapeutic approaches. Among these types, lung cancer is the third most common cancer, which affects all over the population. Lung cancer is a cancer that forms in tissues of the lung, usually in the cells that line the air passages. There are two main types of lung cancer: small cell and non-small cell lung cancer. These two types grow differently and are treated differently. This review explores the application of advanced deep learning (DL) techniques in enhancing biosignature immunotherapy tools for the prediction of immune phenotypes in lung cancer patients. The study systematically analyses recent research integrating multi-modal biomedical data, such as radiomics, genomics, transcriptomics, and histopathological images, to develop robust DL-based predictive models. A well-defined literature search strategy, inclusion/exclusion criteria, and a PRISMA-guided screening process ensure transparency and reproducibility. Emphasis is placed on identifying key predictive biomarkers, including Programmed Death-Ligand 1 (PD-L1) expression, Tumor Mutational Burden (TMB), Microsatellite Instability (MSI), and APOBEC mutational signatures, which are vital for personalizing immunotherapy. The review also incorporates a quality assessment framework to evaluate the methodological rigor of the included studies. Enhanced technical details, such as model architecture, validation strategies, hyperparameter tuning, and standardized performance metrics like AUC-ROC and Harrell's C-index, are presented to facilitate cross-study comparisons. This review underscores the transformative role of DL in precision oncology and highlights the potential for integrating biosignatures into clinical workflows to improve immunotherapy outcomes in lung cancer.

PMID:40447924 | DOI:10.1007/s12672-025-02771-1

Sci Rep. 2025 May 30;15(1):19008. doi: 10.1038/s41598-025-02924-w.

ABSTRACT

Cardiovascular disease (CVD) is one of the severe disorders that requires effectual solutions. CVD mainly affects heart functionality in the human body. The impacts of heart disorders are hazardous, which primarily spread from arrhythmia and higher hypertension to heart attack or stroke and also death. Employing newly established data analysis techniques and inspecting a patient's health record might help recognize CVD promptly. In general, pervasive healthcare (PH) services have the potential to enhance healthcare and the excellence of the lifespan of chronic disease patients over constant monitoring. However, the conventional risk evaluation techniques are neither dynamic nor accurate because they stick to the arithmetical data and ignore the significant time-based effects of the crucial signs. So, recent work has utilized machine learning and deep learning methodologies for predicting CVD on clinical datasets. These methods can decrease death rates by predicting CVD depending on the medical data and the patient's severity level. This manuscript presents a deep convolutional fuzzy neural networks with stork optimization on cardiovascular disease classification (DCFNN-SOCVDC) technique for PH services. The main goal of the DCFNN-SOCVDC method is to detect and classify CVD in the healthcare environment. At first, the presented DCFNN-SOCVDC model performs data preprocessing by utilizing Z-score normalization to preprocess the medical data. For the feature selection process, the presented DCFNN-SOCVDC technique utilizes an arithmetic optimization algorithm model. Besides, the deep convolutional fuzzy neural network (DCFNN) method is employed to identify and classify CVD. Eventually, the presented DCFNN-SOCVDC approach employs a stork optimization algorithm method for the hyperparameter tuning method involved in the DCFNN model. The performance of the DCFNN-SOCVDC approach is evaluated using a CVD dataset, and the results are assessed based on various metrics. The performance validation of the DCFNN-SOCVDC approach portrayed a superior accuracy value of 99.05% over recent models.

PMID:40447750 | DOI:10.1038/s41598-025-02924-w

Sci Rep. 2025 May 30;15(1):19034. doi: 10.1038/s41598-025-04136-8.

ABSTRACT

Breast cancer diagnosis remains a crucial challenge in medical research, necessitating accurate and automated detection methods. This study introduces an advanced deep learning framework for histopathological image classification, integrating AlexNet and Gated Recurrent Unit (GRU) networks, optimized using the Hippopotamus Optimization Algorithm (HOA). Initially, DenseNet-41 extracts intricate spatial features from histopathological images. These features are then processed by the hybrid AlexNet-GRU model, leveraging AlexNet's robust feature extraction and GRU's sequential learning capabilities. HOA is employed to fine-tune hyperparameters, ensuring optimal model performance. The proposed approach is evaluated on benchmark datasets (BreakHis and BACH), achieving a classification accuracy of 99.60%, surpassing existing state-of-the-art models. The results demonstrate the efficacy of integrating deep learning with bio-inspired optimization techniques in breast cancer detection. This research offers a robust and computationally efficient framework for improving early diagnosis and clinical decision-making, potentially enhancing patient outcomes.

PMID:40447726 | DOI:10.1038/s41598-025-04136-8

Sci Rep. 2025 May 30;15(1):19071. doi: 10.1038/s41598-025-02194-6.

ABSTRACT

With advancements in drone control technology, low-altitude remote sensing image processing holds significant potential for intelligent, real-time urban management. However, achieving high accuracy with deep learning algorithms remains challenging due to the stringent requirements for low computational cost, minimal parameters, and real-time performance. This study introduces the Global Object-Oriented Dynamic Network (GOOD-Net) algorithm, comprising three fundamental components: an object-oriented, dynamically adaptive backbone network; a neck network designed to optimize the utilization of global information; and a task-specific processing head augmented for detailed feature refinement. Novel module components, such as the ReSSD Block, GPSA, and DECBS, are integrated to enable fine-grained feature extraction while maintaining computational and parameter efficiency. The efficacy of individual components in the GOOD-Net algorithm, as well as their synergistic interaction, is assessed through ablation experiments. Evaluation conducted on the VisDrone dataset demonstrates substantial enhancements. Furthermore, experiments assessing robustness and deployment on edge devices validate the algorithm's scalability and practical applicability. Visualization methods further highlight the algorithm's performance advantages. This research presents a scalable object detection framework adaptable to various application scenarios and contributes a novel design paradigm for efficient deep learning-based object detection.

PMID:40447715 | DOI:10.1038/s41598-025-02194-6

Sci Rep. 2025 May 30;15(1):19081. doi: 10.1038/s41598-025-02728-y.

ABSTRACT

To address inter-frame motion artifacts in ultrasound quantitative high-definition microvasculature imaging (qHDMI), we introduced a novel deep learning-based motion correction technique. This approach enables the derivation of more accurate quantitative biomarkers from motion-corrected HDMI images, improving the classification of thyroid nodules. Inter-frame motion, often caused by carotid artery pulsation near the thyroid, can degrade image quality and compromise biomarker reliability, potentially leading to misdiagnosis. Our proposed technique compensates for these motion-induced artifacts, preserving the fine vascular structures critical for accurate biomarker extraction. In this study, we utilized the motion-corrected images obtained through this framework to derive the quantitative biomarkers and evaluated their effectiveness in thyroid nodule classification. We segregated the dataset according to the amount of motion into low and high motion containing cases based on the inter-frame correlation values and performed the thyroid nodule classification for the high motion containing cases and the full dataset. A comprehensive analysis of the biomarker distributions obtained after using the corresponding motion-corrected images demonstrates the significant differences between benign and malignant nodule biomarker characteristics compared to the original motion-containing images. Specifically, the bifurcation angle values derived from the quantitative high-definition microvasculature imaging (qHDMI) become more consistent with the usual trend after motion correction. The classification results demonstrated that sensitivity remained unchanged for groups with less motion, while improved by 9.2% for groups with high motion. These findings highlight that motion correction helps in deriving more accurate biomarkers, which improves the overall classification performance.

PMID:40447670 | DOI:10.1038/s41598-025-02728-y

Nat Commun. 2025 May 30;16(1):5037. doi: 10.1038/s41467-025-60434-9.

ABSTRACT

Visualizing high-dimensional data is essential for understanding biomedical data and deep learning models. Neighbor embedding methods, such as t-SNE and UMAP, are widely used but can introduce misleading visual artifacts. We find that the manifold learning interpretations from many prior works are inaccurate and that the misuse stems from a lack of data-independent notions of embedding maps, which project high-dimensional data into a lower-dimensional space. Leveraging the leave-one-out principle, we introduce LOO-map, a framework that extends embedding maps beyond discrete points to the entire input space. We identify two forms of map discontinuity that distort visualizations: one exaggerates cluster separation and the other creates spurious local structures. As a remedy, we develop two types of point-wise diagnostic scores to detect unreliable embedding points and improve hyperparameter selection, which are validated on datasets from computer vision and single-cell omics.

PMID:40447630 | DOI:10.1038/s41467-025-60434-9

Sci Rep. 2025 May 30;15(1):19036. doi: 10.1038/s41598-025-00236-7.

ABSTRACT

Optimal surgical methods require accurate prediction of extraction difficulty and complications. Although various automated methods related to third molar (M3) extraction have been proposed, none fully predict both extraction difficulty and post-extraction complications. This study proposes an automatic diagnosis method based on state-of-the-art semantic segmentation and classification models to predict the extraction difficulty of maxillary and mandibular M3s and possible complications (sinus perforation and inferior alveolar nerve (IAN) injury). A dataset of 4,903 orthopantomographys (OPGs), annotated by experts, was used. The proposed diagnosis method segments M3s (#18, #28, #38, #48), second molars (#17, #27, #37, #47), maxillary sinuses, and inferior alveolar canal (IAC) in OPGs using a segmentation model and extracts the region of interest (RoI). Using the RoI as input, the classification model predicts extraction difficulty and complication possibilities. The model achieved 87.97% and 88.85% accuracy in predicting maxillary and mandibular M3 extraction difficulty, with area under the receiver operating characteristic curve (AUROC) of 96.25% and 97.3%, respectively. It also predicted the possibility of sinus perforation and IAN injury with 91.45% and 88.47% accuracy, and AUROC of 91.78% and 94.13%, respectively. Our results show that the proposed method effectively predicts the extraction difficulty and complications of maxillary and mandibular M3s using OPG, and could serve as a decision support system for clinicians before surgery.

PMID:40447616 | DOI:10.1038/s41598-025-00236-7

Nat Commun. 2025 May 30;16(1):5027. doi: 10.1038/s41467-025-60093-w.

ABSTRACT

The objective of optical super-resolution imaging is to acquire reliable sub-diffraction information on bioprocesses to facilitate scientific discovery. Structured illumination microscopy (SIM) is acknowledged as the optimal modality for live-cell super-resolution imaging. Although recent deep learning techniques have substantially advanced SIM, their transparency and reliability remain uncertain and under-explored, often resulting in unreliable results and biological misinterpretation. Here, we develop Bayesian deep learning (BayesDL) for SIM, which enhances the reconstruction of densely labeled structures while enabling the quantification of super-resolution uncertainty. With the uncertainty, BayesDL-SIM achieves high-fidelity distribution-informed SIM imaging, allowing for the communication of credibility estimates to users regarding the model outcomes. We also demonstrate that BayesDL-SIM boosts SIM reliability by identifying and preventing erroneous generalizations in various model misuse scenarios. Moreover, the BayesDL uncertainty shows versatile utilities for daily super-resolution imaging, such as error estimation, data acquisition evaluation, etc. Furthermore, we demonstrate the effectiveness and superiority of BayesDL-SIM in live-cell imaging, which reliably reveals F-actin dynamics and the reorganization of the cell cytoskeleton. This work lays the foundation for the reliable implementation of deep learning-based SIM methods in practical applications.

PMID:40447610 | DOI:10.1038/s41467-025-60093-w

JMIR Form Res. 2025 May 30;9:e56057. doi: 10.2196/56057.

ABSTRACT

BACKGROUND: Conventional approaches for major depressive disorder (MDD) screening rely on two effective but subjective paradigms: self-rated scales and clinical interviews. Artificial intelligence (AI) can potentially contribute to psychiatry, especially through the use of objective data such as objective audiovisual signals.

OBJECTIVE: This study aimed to evaluate the efficacy of different paradigms using AI analysis on audiovisual signals.

METHODS: We recruited 89 participants (mean age, 37.1 years; male: 30/89, 33.7%; female: 59/89, 66.3%), including 41 patients with MDD and 48 asymptomatic participants. We developed AI models using facial movement, acoustic, and text features extracted from videos obtained via a tool, incorporating four paradigms: conventional scale (CS), question and answering (Q&A), mental imagery description (MID), and video watching (VW). Ablation experiments and 5-fold cross-validation were performed using two AI methods to ascertain the efficacy of paradigm combinations. Attention scores from the deep learning model were calculated and compared with correlation results to assess comprehensibility.

RESULTS: In video clip-based analyses, Q&A outperformed MID with a mean binary sensitivity of 79.06% (95%CI 77.06%-83.35%; P=.03) and an effect size of 1.0. Among individuals, the combination of Q&A and MID outperformed MID alone with a mean extent accuracy of 80.00% (95%CI 65.88%-88.24%; P= .01), with an effect size 0.61. The mean binary accuracy exceeded 76.25% for video clip predictions and 74.12% for individual-level predictions across the two AI methods, with top individual binary accuracy of 94.12%. The features exhibiting high attention scores demonstrated a significant overlap with those that were statistically correlated, including 18 features (all Ps<.05), while also aligning with established nonverbal markers.

CONCLUSIONS: The Q&A paradigm demonstrated higher efficacy than MID, both individually and in combination. Using AI to analyze audiovisual signals across multiple paradigms has the potential to be an effective tool for MDD screening.

PMID:40446148 | DOI:10.2196/56057

PLoS One. 2025 May 30;20(5):e0321559. doi: 10.1371/journal.pone.0321559. eCollection 2025.

ABSTRACT

Deep learning has revolutionized the classification of land cover objects in hyperspectral images (HSIs), particularly by managing the complex 3D cube structure inherent in HSI data. Despite these advances, challenges such as data redundancy, computational costs, insufficient sample sizes, and the curse of dimensionality persist. Traditional 2D Convolutional Neural Networks (CNNs) struggle to fully leverage the interconnections between spectral bands in HSIs, while 3D CNNs, which capture both spatial and spectral features, require more sophisticated design. To address these issues, we propose a novel multilayered, multi-branched 2D-3D CNN model in this paper that integrates Segmented Principal Component Analysis (SPCA) and the minimum-Redundancy-Maximum-Relevance (mRMR) technique. This approach explores the local structure of the data and ranks features by significance. Our approach then hierarchically processes these features: the shallow branch handles the least significant features, the deep branch processes the most critical features, and the mid branch deals with the remaining features. Experimental results demonstrate that our proposed method outperforms most of the state-of-the-art techniques on the Salinas Scene, University of Pavia, and Indian Pines hyperspectral image datasets achieving 100%, 99.94%, and 99.12% Overall Accuracy respectively.

PMID:40446012 | DOI:10.1371/journal.pone.0321559

PLoS One. 2025 May 30;20(5):e0322405. doi: 10.1371/journal.pone.0322405. eCollection 2025.

ABSTRACT

Arsenic contamination of drinking water is a significant health risk. Countries such as Bangladesh's rural areas and regions are in the red alert zone because groundwater is the only primary source of drinking. Early detection of arsenic disease is critical for mitigating long-term health issues. However, these approaches are not widely accepted. In this study, we proposed a fusion approach for the detection of arsenic skin disease. The proposed model is a combination of the Xception model with the Inception module in a deep learning architecture named "ArsenicNet." The model was trained and tested on a publicly available image dataset named "ArsenicSkinImageBD" which contains only 1287 samples and is based on Bangladeshi people. The proposed model achieved the best accuracy through proper experimentation compared to several state-of-the-art deep learning models, including InceptionV3, VGG19, EfficientNetV2B0, ResNet152V2, ViT, and Xception. The proposed model achieved an accuracy of 97.69% and an F1 score of 97.63%, demonstrating superior performance. This research indicates that our proposed model can detect complex patterns in which arsenic skin disease is present, leading to a superior detection performance. Moreover, data augmentation techniques and earlystoping function were used to prevent models overfitting. This study highlights the potential of sophisticated deep learning methodologies to enhance the accuracy of arsenic detection and prevent premature interventions in the diagnosis of arsenic-related illnesses in people. This research contributes to ongoing efforts to develop robust and scalable solutions to monitor and manage arsenic contamination-related health issues.

PMID:40446004 | DOI:10.1371/journal.pone.0322405

PLoS One. 2025 May 30;20(5):e0321770. doi: 10.1371/journal.pone.0321770. eCollection 2025.

ABSTRACT

Accurate diagnosis of apple diseases is vital for tree health, yield improvement, and minimizing economic losses. This study introduces a deep learning-based model to tackle issues like limited datasets, small sample sizes, and low recognition accuracy in detecting apple leaf diseases. The approach begins with enhancing the CycleGAN-M network using a multi-scale attention mechanism to generate synthetic samples, improving model robustness and generalization by mitigating imbalances in disease-type representation. Next, an improved YOLOv8s-KEF model is introduced to overcome limitations in feature extraction, particularly for small lesions and complex textures in natural environments. The model's backbone replaces the standard C2f structure with C2f-KanConv, significantly enhancing disease recognition capabilities. Additionally, we optimize the detection head with Efficient Multi-Scale Convolution (EMS-Conv), improving the model's ability to detect small targets while maintaining robustness and generalization across diverse disease types and conditions. Incorporating Focal-EIoU further reduces missed and false detections, enhancing overall accuracy. The experiment results demonstrate that the YOLOv8s-KEF model achieves 95.0% in accuracy, 93.1% in recall, 95.8% in precision, and an F1-score of 94.5%. Compared to the original YOLOv8s model, the proposed model improves accuracy by 7.2%, precision by 6.5%, and F1-score by 5.0%, with only a modest 6MB increase in model size. Furthermore, compared to Faster RCNN, ResNet50, SSD, YOLOv3-tiny, YOLOv6, YOLOv9s, and YOLOv10m, our model demonstrates substantial improvements, with up to 30.2% higher precision and 18.0% greater accuracy. This study used CycleGAN-M and YOLOv8s-KEF methods to enhance the detection capability of apple leaf diseases.

PMID:40445983 | DOI:10.1371/journal.pone.0321770