PLoS One. 2025 Jul 24;20(7):e0326579. doi: 10.1371/journal.pone.0326579. eCollection 2025.

ABSTRACT

Deep learning models for diagnostic applications require large amounts of sensitive patient data, raising privacy concerns under centralized training paradigms. We propose FedGAN, a federated learning framework for synthetic medical image generation that combines Generative Adversarial Networks (GANs) with cross-silo federated learning. Our approach pretrains a DCGAN on abdominal CT scans and fine-tunes it collaboratively across clinical silos using diabetic retinopathy datasets. By federating the GAN's discriminator and generator via the Federated Averaging (FedAvg) algorithm, FedGAN generates high-quality synthetic retinal images while complying with HIPAA and GDPR. Experiments demonstrate that FedGAN achieves a realism score of 0.43 (measured by a centralized discriminator). This work bridges data scarcity and privacy challenges in medical AI, enabling secure collaboration across institutions.

PMID:40705831 | DOI:10.1371/journal.pone.0326579

PLoS One. 2025 Jul 24;20(7):e0328138. doi: 10.1371/journal.pone.0328138. eCollection 2025.

ABSTRACT

With the acceleration of the global urbanization process, landscape design is facing increasingly complex challenges. Traditional manual design methods are gradually unable to meet the needs for efficiency, precision, and sustainability. To address this issue, this paper proposes an intelligent landscape design generation model based on multimodal deep learning, namely CBS3-LandGen. By integrating image data, text data, and generation optimization techniques, this model can generate landscape plans that meet the design objectives within limited time and resources.Specifically, the model employs the ConvNeXt network to process image data, uses the BART model to analyze text information, and optimizes the generation effect through StyleGAN3. This multimodal architecture enables the model to perform excellently in terms of image generation quality, text generation consistency, and the fusion of images and text. In the experiments, we trained and tested the model using the DeepGlobe and COCO datasets. The results show that on the DeepGlobe dataset, the Frechet Inception Distance (FID) is 25.5 and the Inception Score (IS) is 4.3; on the COCO dataset, the FID is 30.2 and the IS is 4.0. These results demonstrate the superiority of CBS3-LandGen in generation tasks, especially in aspects such as image quality, diversity, and multimodal data fusion. The method proposed in this paper provides new ideas for intelligent landscape design and promotes the integration of landscape design and deep learning technologies. Future research will further optimize the model's performance, improve training efficiency, and expand its application potential in practical landscape design, urban planning, ecological protection, and other fields. The code for implementing CBS3-LandGen Model is available at https://github.com/LMZ81/CBS3-LandGen.git.

PMID:40705822 | DOI:10.1371/journal.pone.0328138

PLoS Comput Biol. 2025 Jul 24;21(7):e1013242. doi: 10.1371/journal.pcbi.1013242. Online ahead of print.

ABSTRACT

Mechanistic models of dynamic, interacting cell populations have yielded many insights into the growth and resolution of immune responses. Historically these models have described the behavior of pre-defined cell types based on small numbers of phenotypic markers. The ubiquity of deep phenotyping therefore presents a new challenge; how do we confront tractable and interpretable mathematical models with high-dimensional data? To tackle this problem, we studied the development and persistence of lung-resident memory CD4 and CD8 T cells ([Formula: see text]) in mice infected with influenza virus. We developed an approach in which dynamical model parameters and the population structure are inferred simultaneously. This method uses deep learning and stochastic variational inference and is trained on the single-cell flow-cytometry data directly, rather than on the kinetics of pre-identified clusters. We show that during the resolution phase of the immune response, memory CD4 and CD8 T cells within the lung are phenotypically diverse, with subsets exhibiting highly distinct and time-dependent dynamics. [Formula: see text] heterogeneity is maintained long-term by ongoing differentiation of relatively persistent Bcl-2hi CD4 and CD8 [Formula: see text] subsets which resolve into distinct functional populations. Our approach yields new insights into the dynamics of tissue-localized immune memory, and is a novel basis for interpreting time series of high-dimensional data, broadly applicable to diverse biological systems.

PMID:40705818 | DOI:10.1371/journal.pcbi.1013242

PLoS One. 2025 Jul 24;20(7):e0328295. doi: 10.1371/journal.pone.0328295. eCollection 2025.

ABSTRACT

The number of commercially available artificial intelligence (AI) tools to support radiological workflows is constantly increasing, yet dedicated solutions for children are largely unavailable. Here, we repurposed an AI-tool developed for chest radiograph interpretation in adults (Lunit INSIGHT CXR) and investigated its diagnostic performance in a real-world pediatric clinical dataset. 958 consecutive frontal chest radiographs of children aged 2-14 years were included and analyzed with the commercially available AI-tool. The reference standard was determined in a dedicated reading session by a board-certified radiologist. The original reports validated by specialized pediatric radiologists, were considered as second readings. All discordant findings were reanalyzed in consensus. The diagnostic performance of the AI-tool was validated using standard measures of accuracy. For this, the continuous AI output (ranging from 0-100) was binarized using vendor recommended thresholds recommended for adults and optimized thresholds identified for children. Relevant findings were defined as consolidation, atelectasis, nodule, cardiomegaly, mediastinal widening due to mass, pleural effusion and pneumothorax. 200 radiographs [20.9%] demonstrated at least one relevant pathology. Using the adult threshold, the AI-tool showed a high performance for all relevant findings with an AUC 0.94 (95% CI: 0.92-0.95) and. In stratified analysis by age (2-7 vs. 7-14-years-old) a significantly higher performance (p < 0.001) was found for older children with an AUC of 0.96 (95% CI: 0.94-0.98) with a sensitivity and specificity of 87.5% and 82.3% respectively, which further increased using optimized thresholds for children. Repurposing existing AI-tools developed for adult application to pediatric patients could support clinical workflows until dedicated solutions become available.

PMID:40705715 | DOI:10.1371/journal.pone.0328295

IEEE Trans Med Imaging. 2025 Jul 24;PP. doi: 10.1109/TMI.2025.3591820. Online ahead of print.

ABSTRACT

Doppler ultrasound modalities, which include spectral Doppler and color flow imaging, are frequently used tools for flow diagnostics because of their real-time point-of-care applicability and high temporal resolution. When implemented using pulse-echo sensing and phase shift estimation principles, this modality's pulse repetition frequency (PRF) is known to influence the maximum detectable velocity. If the PRF is inevitably set below the Nyquist limit due to imaging requirements or hardware constraints, aliasing errors or spectral overlap may corrupt the estimated flow data. To solve this issue, we have devised a deep learning-based framework, powered by a custom slow-time upsampling network (SUP-Net) that leverages spatiotemporal characteristics to upsample the received ultrasound signals across pulse echoes acquired using high-frame-rate ultrasound (HiFRUS). Our framework infers high-PRF signals from signals acquired at low PRF, thereby improving Doppler ultrasound's flow estimation quality. SUP-Net was trained and evaluated on in vivo femoral acquisitions from 20 participants and was applied recursively to resolve scenarios with excessive aliasing across a range of PRFs. We report the successful reconstruction of slow-time signals with frequency content that exceeds the Nyquist limit once and twice. By operating on the fundamental slow-time signals, our framework can resolve aliasing-related artifacts in several downstream modalities, including color Doppler and pulse wave Doppler.

PMID:40705591 | DOI:10.1109/TMI.2025.3591820

Proc Natl Acad Sci U S A. 2025 Jul 29;122(30):e2424662122. doi: 10.1073/pnas.2424662122. Epub 2025 Jul 24.

ABSTRACT

We analyze changes in pedestrian behavior over a 30-y period in four urban public spaces located in New York, Boston, and Philadelphia. Building on William Whyte's observational work, which involved manual video analysis of pedestrian behaviors, we employ computer vision and deep learning techniques to examine video footage from 1979-80 and 2008-10. Our analysis measures changes in walking speed, lingering behavior, group sizes, and group formation. We find that the average walking speed has increased by 15%, while the time spent lingering in these spaces has halved across all locations. Although the percentage of pedestrians walking alone remained relatively stable (from 67% to 68%), the frequency of group encounters declined, indicating fewer interactions in public spaces. This shift suggests that urban residents are using streets as thoroughfares rather than as social spaces, which has important implications for the role of public spaces in fostering social engagement.

PMID:40705424 | DOI:10.1073/pnas.2424662122

EJNMMI Phys. 2025 Jul 24;12(1):72. doi: 10.1186/s40658-025-00785-w.

ABSTRACT

BACKGROUND: Single-photon emission computed tomography (SPECT) plays a crucial role in detecting bone metastases from lung cancer. However, its low spatial resolution and lesion similarity to benign structures present significant challenges for accurate segmentation, especially for lesions of varying sizes.

METHODS: We propose a deep learning-based segmentation framework that integrates conditional adversarial learning with a multi-scale feature extraction generator. The generator employs cascade dilated convolutions, multi-scale modules, and deep supervision, while the discriminator utilizes multi-scale L1 loss computed on image-mask pairs to guide segmentation learning.

RESULTS: The proposed model was evaluated on a dataset of 286 clinically annotated SPECT scintigrams. It achieved a Dice Similarity Coefficient (DSC) of 0.6671, precision of 0.7228, and recall of 0.6196 - outperforming both classical and recent adversarial segmentation models in multi-scale lesion detection, especially for small and clustered lesions.

CONCLUSION: Our results demonstrate that the integration of multi-scale feature learning with adversarial supervision significantly improves the segmentation of bone metastasis in SPECT imaging. This approach shows potential for clinical decision support in the management of lung cancer.

PMID:40705118 | DOI:10.1186/s40658-025-00785-w

Ecol Lett. 2025 Jul;28(7):e70177. doi: 10.1111/ele.70177.

ABSTRACT

Species Distribution Models based on Convolutional Neural Networks (CNN-SDMs) have recently emerged, demonstrating greater effectiveness than traditional SDMs in several contexts. A limited number of studies, however, have focused on species abundance patterns, as the datasets available for this purpose are generally too small to effectively learn a deep learning model with millions of parameters. Our study demonstrated that CNN-SDMs can circumvent the small sample size of species abundance datasets through the combined use of a large presence-only species dataset and transfer learning to significantly improve the performance of abundance-based CNN-SDMs. Applied to Mediterranean coastal fishes, our approach significantly improves the abundance prediction performance of CNN-SDMs, with average gains of 35% (D-squared regression score). This allows CNN-SDMs to perform better than classical SDMs in abundance prediction, with average gains of 10%. These gains are stemming from enhanced abundance predictions for rare species and where widespread species are locally rare.

PMID:40704696 | DOI:10.1111/ele.70177

Orthod Craniofac Res. 2025 Jul 24. doi: 10.1111/ocr.70008. Online ahead of print.

ABSTRACT

OBJECTIVE: The study aims to evaluate the effectiveness of deep learning algorithms in skeletal age estimation by comparing the diagnostic reliability of five different hand-wrist maturation (HWM) assessment methods.

MATERIALS AND METHODS: A total of 6572 hand-wrist radiographs from orthodontic patients aged 8-16 years were retrospectively analysed. Radiographs were categorised into five groups based on HWM classification methods: (I) Björk's nine-stage, (II) Hägg and Taranger's five-stage, (III) Chapman's four-stage, (IV) three-stage hook of hamate ossification based and (V) simplified three-stage Björk's classification based. YOLOv8x-based deep learning models were trained separately for each group. The dataset was split into training, validation and test subsets. Performance was evaluated using accuracy, precision, recall, F1 score and AUC metrics.

RESULTS: The YOLOv8x-cls model demonstrated high classification performance across all five groups. Group IV and Group II achieved the highest accuracy and F1 scores, with average F1 values of 0.99 and 0.96, respectively. Group III and Group V also showed strong performance (F1 = 0.93 and 0.92). In Group I, slightly lower classification performance was observed in the S-H2 and MP3-Cap stages (F1 = 0.72-0.74), which correspond to the pubertal growth peak, while early and late skeletal maturation stages were classified with high accuracy and F1 scores. ROC curve analysis further supported these findings, with AUC values for MP3-Cap and S-H2 recorded as 0.70 and 0.75, respectively, whereas higher AUC values were achieved in most other stages across all groups.

CONCLUSION: Deep learning models proved effective in evaluating skeletal maturation across five different HWM methods. Particularly high performance was observed in anatomically distinct regions such as the MP3, adductor sesamoid and hamate bone, which can be reliably identified by general dentists, enabling earlier referrals and timely orthodontic interventions.

PMID:40704688 | DOI:10.1111/ocr.70008

Mov Disord. 2025 Jul 24. doi: 10.1002/mds.30300. Online ahead of print.

ABSTRACT

BACKGROUND: The differentiation between multiple system atrophy (MSA) and Parkinson's disease (PD) based on clinical diagnostic criteria can be challenging, especially at an early stage. Leveraging deep learning methods and magnetic resonance imaging (MRI) data has shown great potential in aiding automatic diagnosis.

OBJECTIVE: The aim was to determine the feasibility of a three-dimensional convolutional neural network (3D CNN)-based approach using multimodal, multicentric MRI data for differentiating MSA and its variants from PD.

METHODS: MRI data were retrospectively collected from three MSA French reference centers. We computed quantitative maps of gray matter density (GD) from a T1-weighted sequence and mean diffusivity (MD) from diffusion tensor imaging. These maps were used as input to a 3D CNN, either individually ("monomodal," "GD" or "MD") or in combination ("bimodal," "GD-MD"). Classification tasks included the differentiation of PD and MSA patients. Model interpretability was investigated by analyzing misclassified patients and providing a visual interpretation of the most activated regions in CNN predictions.

RESULTS: The study population included 92 patients with MSA (50 with MSA-P, parkinsonian variant; 33 with MSA-C, cerebellar variant; 9 with MSA-PC, mixed variant) and 64 with PD. The best accuracies were obtained for the PD/MSA (0.88 ± 0.03 with GD-MD), PD/MSA-C&PC (0.84 ± 0.08 with MD), and PD/MSA-P (0.78 ± 0.09 with GD) tasks. Patients misclassified by the CNN exhibited fewer and milder image alterations, as found using an image-based z score analysis. Activation maps highlighted regions involved in MSA pathophysiology, namely the putamen and cerebellum.

CONCLUSIONS: Our findings hold promise for developing an efficient, MRI-based, and user-independent diagnostic tool suitable for differentiating parkinsonian syndromes in clinical practice. © 2025 The Author(s). Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

PMID:40704399 | DOI:10.1002/mds.30300

Front Big Data. 2025 Jul 9;8:1569147. doi: 10.3389/fdata.2025.1569147. eCollection 2025.

ABSTRACT

INTRODUCTION: The widespread emergence of deepfake videos presents substantial challenges to the security and authenticity of digital content, necessitating robust detection methods. Deepfake detection remains challenging due to the increasing sophistication of forgery techniques. While existing methods often focus on spatial features, they may overlook crucial temporal information distinguishing real from fake content and need to investigate several other Convolutional Neural Network architectures on video-based deep fake datasets.

METHODS: This study introduces an RLNet deep learning framework that utilizes ResNet and Long Short Term Memory (LSTM) networks for high-precision deepfake video detection. The key objective is exploiting spatial and temporal features to discern manipulated content accurately. The proposed approach starts with preprocessing a diverse dataset with authentic and deepfake videos. The ResNet component captures intricate spatial anomalies at the frame level, identifying subtle manipulations. Concurrently, the LSTM network analyzes temporal inconsistencies across video sequences, detecting dynamic irregularities that signify deepfake content.

RESULTS AND DISCUSSION: Experimental results demonstrate the effectiveness of the combined ResNet and LSTM approach, showing an accuracy of 95.2% and superior detection capabilities compared to existing methods like EfficientNet and Recurrent Neural Networks (RNN). The framework's ability to handle various deepfake techniques and compression levels highlights its versatility and robustness. This research significantly contributes to digital media forensics by providing an advanced tool for detecting deepfake videos, enhancing digital content's security and integrity. The efficacy and resilience of the proposed system are evidenced by deepfake detection, while our visualization-based interpretability provides insights into our model.

PMID:40704217 | PMC:PMC12283977 | DOI:10.3389/fdata.2025.1569147

Ewha Med J. 2025 Jan;48(1):e6. doi: 10.12771/emj.2025.e6. Epub 2025 Jan 31.

ABSTRACT

Shoulder diseases pose a significant health challenge for older adults, often causing pain, functional decline, and decreased independence. This narrative review explores how deep learning (DL) can address diagnostic challenges by automating tasks such as image segmentation, disease detection, and motion analysis. Recent research highlights the effectiveness of DL-based convolutional neural networks and machine learning frameworks in diagnosing various shoulder pathologies. Automated image analysis facilitates the accurate assessment of rotator cuff tear size, muscle degeneration, and fatty infiltration in MRI or CT scans, frequently matching or surpassing the accuracy of human experts. Convolutional neural network-based systems are also adept at classifying fractures and joint conditions, enabling the rapid identification of common causes of shoulder pain from plain radiographs. Furthermore, advanced techniques like pose estimation provide precise measurements of the shoulder joint's range of motion and support personalized rehabilitation plans. These automated approaches have also been successful in quantifying local osteoporosis, utilizing machine learning-derived indices to classify bone density status. DL has demonstrated significant potential to improve diagnostic accuracy, efficiency, and consistency in the management of shoulder diseases in older patients. Machine learning-based assessments of imaging data and motion parameters can help clinicians optimize treatment plans and improve patient outcomes. However, to ensure their generalizability, reproducibility, and effective integration into routine clinical workflows, large-scale, prospective validation studies are necessary. As data availability and computational resources increase, the ongoing development of DL-driven applications is expected to further advance and personalize musculoskeletal care, benefiting both healthcare providers and the aging population.

PMID:40704206 | PMC:PMC12277901 | DOI:10.12771/emj.2025.e6

MethodsX. 2025 Jul 10;15:103499. doi: 10.1016/j.mex.2025.103499. eCollection 2025 Dec.

ABSTRACT

The exponential growth of Internet of Things (abbreviated as IoT) has led to a surge in cyber threats, especially botnet attacks that compromise network security. Although machine learning (abbreviated ML) & deep learning (abbreviated as DL) approaches have shown promise in detecting these attacks, they often struggle with limited accuracy & high computational requirements, making them unsuitable for real-time detection in resource-constrained IoT environments. To overcome these limitations, this research proposes an enhanced detection framework based on an improved SqueezeNet model integrated with a Deep Convolutional Neural Network (abbreviated as DCNN) and an optimized stochastic mixed Lp layer. This model aims to improve detection accuracy while maintaining computational efficiency. Experimental evaluation using a large-scale intrusion detection dataset demonstrates that the proposed model significantly outperforms existing techniques such as Bi-GRU, CNN, PolyNet, and LinkNet, achieving a classification accuracy of 0.97 and a reduced false positive rate of 0.054. The complete research process is outlined below:•Data Pre-processing: Min-max normalization is applied to the input dataset to ensure consistent data scaling and enhance model learning performance.•Feature Extraction and Classification: The improved SqueezeNet is integrated with DCNN & a stochastic mixed Lp layer to extract meaningful features and classify attacks accurately.•Model Evaluation: Performance is validated through accuracy, precision, recall, and false positive rate using a benchmark intrusion detection dataset.

PMID:40704174 | PMC:PMC12284568 | DOI:10.1016/j.mex.2025.103499

JHEP Rep. 2025 May 30;7(8):101465. doi: 10.1016/j.jhepr.2025.101465. eCollection 2025 Aug.

ABSTRACT

BACKGROUND & AIMS: Liver regeneration is essential for recovery following injury, but this process can be impaired by factors such as sex, age, metabolic disorders, fibrosis, and immunosuppressive therapies. We aimed to identify key transcriptomic, proteomic, and serum biomarkers of regeneration in mouse models under these diverse conditions using systems biology and machine learning approaches.

METHODS: Six mouse models, each undergoing 75% hepatectomy, were used to study regeneration across distinct clinical contexts: young males and females, aged mice, stage 2 fibrosis, steatosis, and tacrolimus exposure. A novel contrastive deep learning framework with triplet loss was developed to map regenerative trajectories and identify genes associated with regenerative efficiency.

RESULTS: Despite achieving ≥75% liver mass restoration by day 7, regeneration was significantly delayed in aged, steatotic, and fibrotic models, as indicated by reduced Ki-67 staining on day 2 (p <0.0001 for all). Interestingly, fibrotic livers exhibited reduced collagen deposition and partial regression to stage 1 fibrosis post-hepatectomy. Transcriptomic and proteomic analyses revealed consistent downregulation of cell cycle genes in impaired regeneration. The deep learning model integrating clinical and transcriptomic data predicted regenerative outcomes with 87.9% accuracy. SHAP (SHapley Additive exPlanations) highlighted six key predictive genes: Wee1, Rbl1, Gnl3, Mdm2, Cdk2, and Ccne2. Proteomic validation and human SPLiT-seq (split-pool ligation-based transcriptome sequencing) data further supported their relevance across species.

CONCLUSIONS: This study identifies conserved cell cycle regulators underlying efficient liver regeneration and provides a predictive framework for evaluating regenerative capacity. The integration of deep learning and multi-omics profiling provides a promising approach to better understand liver regeneration and may help guide therapeutic strategies, especially in complex clinical settings.

IMPACT AND IMPLICATIONS: The aim of this study was to identify key transcriptomic, proteomic, and serum biomarkers of regeneration in mouse models under diverse conditions, using systems biology and machine learning approaches. Key molecular drivers of liver regeneration across diverse clinical conditions were identified using innovative deep learning and multi-omics approaches. By identifying conserved cell cycle genes predictive of regenerative outcomes, this study offers a powerful framework to assess and potentially enhance liver recovery in older patients, those with fibrosis or steatosis, and/or those under immunosuppression.

PMID:40704068 | PMC:PMC12284365 | DOI:10.1016/j.jhepr.2025.101465

J Pathol Inform. 2025 Jul 1;18:100458. doi: 10.1016/j.jpi.2025.100458. eCollection 2025 Aug.

ABSTRACT

Digital pathology is an emerging field that is gaining popularity due to its numerous advantages over traditional pathology methods. Digital pathology allows for the remote examination of tissue samples, increasing efficiency and reducing costs. The field of digital pathology is experiencing a boom of data, creating space for new tools to be implemented that have not been used in pathology prior. Benford's Law is a statistical tool commonly used to analyze large datasets by other top organizations. Benford's Law is a law of frequency of first and second digits and whether they would appear normally in nature. With research in multiple fields of medicine moving into a digital era, tools that had once been used elsewhere to analyze digital images could translate well into pathology. Quantitative histomorphometry is a tool in digital pathology that analyzes digital images and collects morphological and histological data of whole-slide images, with more techniques being developed in digital pathology, such as deep learning, creating a more accurate 3D analysis of the cell. Easy and quick tools are needed to analyze the large datasets that are being extracted quickly. We believe that Benford's Law is a statistical outlook that can be easily implemented for similar use in whole-slide image analysis. When a system is disrupted by disease, it will distort the normal, natural growth of cells throughout the organ. Open access tools such as QuPath have created a way to obtain categories of data to analyze, such as the size of a cell or the amount of staining it absorbs. Slides of normal liver cells were collected and compared to slides of a liver with cancer. The liver was selected because of its well-demarcated cytoplastic borders and nucleus. A total of 25 liver tissue slides were collected. The graph of naturalness is compared to analyze ways to detect variability between normal liver cells and cancer liver cells. 206,700 cells from 15 slides of 7 cancer patients' liver tissue samples (15 slides total) and 116,339 cells from 5 slides of normal liver tissue were collected, totaling 323,039 cells from 20 slides. Of the seven cancer patients, five were previously diagnosed with cholangiocarcinoma, and two were diagnosed with adenomas/adenocarcinoma. The study found that of the 13 data categories provided by QuPath, such as cell size, nucleus size, and color absorbance, two met the Chi-square goodness of fit (χ2) criteria compared to Benford's Law of Naturalness, providing the most significant feedback. Due to QuPath's inability to distinguish all cytoplastic borders accurately, categories that depict size measurements were not used. Of the two categories that did correlate, such as those that used stain absorbance, 62.5% of slides that exceeded the critical value contained cells of someone diagnosed with cancer. In contrast, all normal slides showed a very low variance. All slides from a cancer patient showed a test statistic above 6 points, whereas the normal tissue slides showed a test statistic below 1.5, strongly correlating with Benford's Law.

PMID:40704058 | PMC:PMC12284048 | DOI:10.1016/j.jpi.2025.100458

Ewha Med J. 2024 Oct;47(4):e71. doi: 10.12771/emj.2024.e71. Epub 2024 Oct 31.

NO ABSTRACT

PMID:40703983 | PMC:PMC12093575 | DOI:10.12771/emj.2024.e71

Front Oral Health. 2025 Jul 9;6:1650353. doi: 10.3389/froh.2025.1650353. eCollection 2025.

NO ABSTRACT

PMID:40703978 | PMC:PMC12283974 | DOI:10.3389/froh.2025.1650353

Front Plant Sci. 2025 Jul 9;16:1567131. doi: 10.3389/fpls.2025.1567131. eCollection 2025.

ABSTRACT

Pinus yunnanensis is a significant tree species in southwest China, crucial for the ecological environment and forest resources. Accurate modeling of its crown profile is essential for forest management and ecological analysis. However, existing modeling approaches face limitations in capturing the crown's spatial heterogeneity and vertical structure. This study aims to propose a novel approach that combines deep learning with a crown competition index to overcome the limitations of traditional models in capturing crown asymmetry and vertical structure, thereby enhancing prediction accuracy. Thus, we developed a hybrid CNN-LSTM-Attention deep learning model combined with a novel Crown Profile Competition Index (CPCI), based on data collected from 629 trees across five age-stratified permanent plots on Cangshan Mountain, Dali, Yunnan Province. Experimental results showed that the hybrid CNN-LSTM and CNN-LSTM-Attention models significantly outperformed the Vanilla LSTM model. In particular, the CNN-LSTM-Attention model achieved the best performance (MSE=0.00755 m2, RMSE=0.08691 m, MAE=0.05198 m, R²=0.98161), with absolute R² improvements of 0.16 and 0.17 over the Vanilla LSTM model by the CNN-LSTM and CNN-LSTM-Attention models, respectively. Additionally, the CNN-LSTM-Attention model demonstrated superior stability and performance in handling directional crown profile datasets. Incorporating CPCI improved prediction accuracy across all models, especially benefiting the Vanilla LSTM model. In conclusion, the proposed hybrid deep learning framework significantly enhances crown profile prediction for Pinus yunnanensis, and the introduction of CPCI provides a more precise representation of vertical and directional crown competition. This improvement facilitates more accurate assessment of tree crown dynamics, which is critical for understanding forest structure and competition.

PMID:40703862 | PMC:PMC12283653 | DOI:10.3389/fpls.2025.1567131

Ewha Med J. 2024 Apr;47(2):e22. doi: 10.12771/emj.2024.e22. Epub 2024 Apr 30.

ABSTRACT

The capabilities of artificial intelligence (AI) have recently surged, largely due to advancements in deep learning inspired by the structure and function of the neural networks of the human brain. In the medical field, the impact of AI spans from diagnostics and treatment recommendations to patient engagement and monitoring, considerably improving efficiency and outcomes. The clinical integration of AI has also been examined in specialties, including pathology, radiology, and oncology. General surgery primarily involves manual manipulation and includes preoperative, intraoperative, and postoperative care, all of which are critical for saving lives. Other fields have strived to utilize and adopt AI; nonetheless, general surgery appears to have retrogressed. In this review, we analyzed the published research, to understand how the application of AI in general surgery differs from that in other medical fields. Based on previous research in other fields, the application of AI in the preoperative stage is nearing feasibility. Ongoing research efforts aim to utilize AI to improve and predict operative outcomes, enhance performance, and improve patient care. However, the use of AI in the operating room remains significantly understudied. Moreover, ethical responsibilities are associated with such research, necessitating extensive work to gather evidence. By fostering interdisciplinary collaboration and leveraging lessons from AI success stories in other fields, AI tools could be specifically tailored for general surgery. Surgeons should be prepared for the integration of AI into clinical practice to achieve better outcomes; therefore, the time has come to consider ethical and legal implications.

PMID:40703691 | PMC:PMC12093534 | DOI:10.12771/emj.2024.e22

Biomed Eng Online. 2025 Jul 23;24(1):92. doi: 10.1186/s12938-025-01426-0.

ABSTRACT

BACKGROUND: The field of contactless health monitoring has witnessed significant advancements with the advent of piezoelectric sensing technology, which enables the monitoring of vital signs such as heart rate and respiration without requiring direct contact with the subject. This is especially advantageous for home sleep monitoring, where traditional wearable devices may be intrusive. However, the acquisition of piezoelectric signals is often impeded by motion artifacts, which are distortions caused by the subject of movements and can obscure the underlying physiological signals. These artifacts can significantly impair the reliability of signal analysis, necessitating effective identification and mitigation strategies. Various methods, including filtering techniques and machine learning approaches, have been employed to address this issue, but the challenge persists due to the complexity and variability of motion artifacts.

METHODS: This study introduces a hybrid model for detecting motion artifacts in ballistocardiogram (BCG) signals, utilizing a dual-channel approach. The first channel uses a deep learning model, specifically a temporal Bidirectional Gated Recurrent Unit combined with a Fully Convolutional Network (BiGRU-FCN), to identify motion artifacts. The second channel employs multi-scale standard deviation empirical thresholds to detect motion. The model was designed to address the randomness and complexity of motion artifacts by integrating deep learning capabilities with manual feature judgment. The data used for this study were collected from patients with sleep apnea using piezoelectric sensors, and the model's performance was evaluated using a set of predefined metrics.

RESULTS: This paper proposes and confirms through analysis that the proposed hybrid model exhibits exceptional accuracy in detecting motion artifacts in ballistocardiogram (BCG) signals. Employing a dual-channel approach, the model integrates multi-scale feature judgment with a BiGRU-FCN deep learning model. It achieved a classification accuracy of 98.61% and incurred only a 4.61% loss of valid signals in non-motion intervals. When tested on data from ten patients with sleep apnea, the model demonstrated robust performance, highlighting its potential for practical use in home sleep monitoring.

CONCLUSION: The proposed hybrid model presents a significant advancement in the detection of motion artifacts in BCG signals. Compared to existing methods such as the Alivar method [29], Enayati method [22], and Wiard method [20], our hybrid model achieves higher classification accuracy (98.61%) and lower valid signal loss ratio (4.61%). This demonstrates the effectiveness of integrating multi-scale standard deviation empirical thresholds with a deep learning model in enhancing the accuracy and robustness of motion artifact detection. This approach is particularly effective for home sleep monitoring, where motion artifacts can significantly impact the reliability of health monitoring data. The study findings suggest that the proposed hybrid model could serve as a valuable tool for improving the accuracy of motion artifact detection in various health monitoring applications.

PMID:40702570 | DOI:10.1186/s12938-025-01426-0