Front Med (Lausanne). 2025 Mar 19;12:1505530. doi: 10.3389/fmed.2025.1505530. eCollection 2025.

ABSTRACT

BACKGROUND: Macular edema (ME) is an ophthalmic disease that poses a serious threat to human vision. Anti-vascular endothelial growth factor (anti-VEGF) therapy has become the first-line treatment for ME due to its safety and high efficacy. However, there are still cases of refractory macular edema and non-responding patients. Therefore, it is crucial to develop automated and efficient methods for predicting therapeutic outcomes.

METHODS: We have developed a predictive model for the surgical efficacy in ME patients based on deep learning and optical coherence tomography (OCT) imaging, aimed at predicting the treatment outcomes at different time points. This model innovatively introduces group convolution and multiple convolutional kernels to handle multidimensional features based on traditional attention mechanisms for visual recognition tasks, while utilizing spatial pyramid pooling (SPP) to combine and extract the most useful features. Additionally, the model uses ResNet50 as a pre-trained model, integrating multiple knowledge through model fusion.

RESULTS: Our proposed model demonstrated the best performance across various experiments. In the ablation study, the model achieved an F1 score of 0.9937, an MCC of 0.7653, an AUC of 0.9928, and an ACC of 0.9877 in the test conducted on the first day after surgery. In comparison experiments, the ACC of our model was 0.9930 and 0.9915 in the first and the third months post-surgery, respectively, with AUC values of 0.9998 and 0.9996, significantly outperforming other models. In conclusion, our model consistently exhibited superior performance in predicting outcomes at various time points, validating its excellence in processing OCT images and predicting postoperative efficacy.

CONCLUSION: Through precise prediction of the response to anti-VEGF therapy in ME patients, deep learning technology provides a revolutionary tool for the treatment of ophthalmic diseases, significantly enhancing treatment outcomes and improving patients' quality of life.

PMID:40177270 | PMC:PMC11961644 | DOI:10.3389/fmed.2025.1505530

Front Oncol. 2025 Mar 19;15:1468232. doi: 10.3389/fonc.2025.1468232. eCollection 2025.

ABSTRACT

OBJECTIVES: Radiotherapy is a fundamental cancer treatment method, and pre-treatment patient-specific quality assurance (prePSQA) plays a crucial role in ensuring dose accuracy and patient safety. Artificial intelligence model for measurement-free prePSQA have been investigated over the last few years. While these models stack successive pooling layers to carry out sequential learning, directly splice together different modalities along channel dimensions and feed them into shared encoder-decoder network, which greatly reduces the anatomical features specific to different modalities. Furthermore, the existing models simply take advantage of low-dimensional dosimetry information, meaning that the spatial features about the complex dose distribution may be lost and limiting the predictive power of the models. The purpose of this study is to develop a novel deep learning model for measurement-guided therapeutic-dose (MDose) prediction from head and neck cancer radiotherapy data.

METHODS: The enrolled 310 patients underwent volumetric-modulated arc radiotherapy (VMAT) were randomly divided into the training set (186 cases, 60%), validation set (62 cases, 20%), and test set (62 cases, 20%). The effective prediction model explicitly integrates the multi-scale features that are specific to CT and dose images, takes into account the useful spatial dose information and fully exploits the mutual promotion within the different modalities. It enables medical physicists to analyze the detailed locations of spatial dose differences and to simultaneously generate clinically applicable dose-volume histograms (DVHs) metrics and gamma passing rate (GPR) outcomes.

RESULTS: The proposed model achieved better performance of MDose prediction, and dosimetric congruence of DVHs, GPR with the ground truth compared with several state-of-the-art models. Quantitative experimental predictions show that the proposed model achieved the lowest values for the mean absolute error (37.99) and root mean square error (4.916), and the highest values for the peak signal-to-noise ratio (52.622), structural similarity (0.986) and universal quality index (0.932). The predicted dose values of all voxels were within 6 Gy in the dose difference maps, except for the areas near the skin or thermoplastic mask indentation boundaries.

CONCLUSIONS: We have developed a feasible MDose prediction model that could potentially improve the efficiency and accuracy of prePSQA for head and neck cancer radiotherapy, providing a boost for clinical adaptive radiotherapy.

PMID:40177241 | PMC:PMC11961879 | DOI:10.3389/fonc.2025.1468232

Front Robot AI. 2025 Mar 19;12:1520374. doi: 10.3389/frobt.2025.1520374. eCollection 2025.

ABSTRACT

A significant number of individuals have been affected by pandemic diseases, such as COVID-19 and seasonal influenza. Nucleic acid testing is a common method for identifying infected patients. However, manual sampling methods require the involvement of numerous healthcare professionals. To address this challenge, we propose a novel transoral swab sampling robot designed to autonomously perform nucleic acid sampling using a visual-tactile fusion approach. The robot comprises a series-parallel hybrid flexible mechanism for precise distal posture adjustment and a visual-tactile perception module for navigation within the subject's oral cavity. The series-parallel hybrid mechanism, driven by flexible shafts, enables omnidirectional bending through coordinated movement of the two segments of the bendable joint. The visual-tactile perception module incorporates a camera to capture oral images of the subject and recognize the nucleic acid sampling point using a deep learning method. Additionally, a force sensor positioned at the distal end of the robot provides feedback on contact force as the swab is inserted into the subject's oral cavity. The sampling robot is capable of autonomously performing transoral swab sampling while navigating using the visual-tactile perception algorithm. Preliminary experimental trials indicate that the designed robot system is feasible, safe, and accurate for sample collection from subjects.

PMID:40177224 | PMC:PMC11961991 | DOI:10.3389/frobt.2025.1520374

Exp Biol Med (Maywood). 2025 Mar 19;250:10359. doi: 10.3389/ebm.2025.10359. eCollection 2025.

ABSTRACT

Opioids exert their analgesic effect by binding to the µ opioid receptor (MOR), which initiates a downstream signaling pathway, eventually inhibiting pain transmission in the spinal cord. However, current opioids are addictive, often leading to overdose contributing to the opioid crisis in the United States. Therefore, understanding the structure-activity relationship between MOR and its ligands is essential for predicting MOR binding of chemicals, which could assist in the development of non-addictive or less-addictive opioid analgesics. This study aimed to develop machine learning and deep learning models for predicting MOR binding activity of chemicals. Chemicals with MOR binding activity data were first curated from public databases and the literature. Molecular descriptors of the curated chemicals were calculated using software Mold2. The chemicals were then split into training and external validation datasets. Random forest, k-nearest neighbors, support vector machine, multi-layer perceptron, and long short-term memory models were developed and evaluated using 5-fold cross-validations and external validations, resulting in Matthews correlation coefficients of 0.528-0.654 and 0.408, respectively. Furthermore, prediction confidence and applicability domain analyses highlighted their importance to the models' applicability. Our results suggest that the developed models could be useful for identifying MOR binders, potentially aiding in the development of non-addictive or less-addictive drugs targeting MOR.

PMID:40177220 | PMC:PMC11961360 | DOI:10.3389/ebm.2025.10359

World J Hepatol. 2025 Mar 27;17(3):101721. doi: 10.4254/wjh.v17.i3.101721.

ABSTRACT

BACKGROUND: In recent years, the utilization of artificial intelligence (AI) technology has gained prominence in the field of liver disease.

AIM: To analyzes AI research in the field of liver disease, summarizes the current research status and identifies hot spots.

METHODS: We searched the Web of Science Core Collection database for all articles and reviews on hepatopathy and AI. The time spans from January 2007 to August 2023. We included 4051 studies for further collection of information, including authors, countries, institutions, publication years, keywords and references. VOS viewer, CiteSpace, R 4.3.1 and Scimago Graphica were used to visualize the results.

RESULTS: A total of 4051 articles were analyzed. China was the leading contributor, with 1568 publications, while the United States had the most international collaborations. The most productive institutions and journals were the Chinese Academy of Sciences and Frontiers in Oncology. Keywords co-occurrence analysis can be roughly summarized into four clusters: Risk prediction, diagnosis, treatment and prognosis of liver diseases. "Machine learning", "deep learning", "convolutional neural network", "CT", and "microvascular infiltration" have been popular research topics in recent years.

CONCLUSION: AI is widely applied in the risk assessment, diagnosis, treatment, and prognosis of liver diseases, with a shift from invasive to noninvasive treatment approaches.

PMID:40177211 | PMC:PMC11959664 | DOI:10.4254/wjh.v17.i3.101721

J Med Imaging (Bellingham). 2025 Mar;12(2):024004. doi: 10.1117/1.JMI.12.2.024004. Epub 2025 Apr 1.

ABSTRACT

PURPOSE: The issue of delivered doses during computed tomography (CT) scans encouraged sparser sets of X-ray projection, severely degrading reconstructions from conventional methods. Although most deep learning approaches benefit from large supervised datasets, they cannot generalize to new acquisition protocols (geometry, source/detector specifications). To address this issue, we developed a method working without training data and independently of experimental setups. In addition, our model may be initialized on small unsupervised datasets to enhance reconstructions.

APPROACH: We propose a conditioned generative latent optimization (cGLO) in which a decoder reconstructs multiple slices simultaneously with a shared objective. It is tested on full-dose sparse-view CT for varying projection sets: (a) without training data against Deep Image Prior (DIP) and (b) with training datasets of multiple sizes against state-of-the-art score-based generative models (SGMs). Peak signal-to-noise ratio (PSNR) and structural SIMilarity (SSIM) metrics are used to quantify reconstruction quality.

RESULTS: cGLO demonstrates better SSIM than SGMs (between + 0.034 and + 0.139 ) and has an increasing advantage for smaller datasets reaching a + 6.06 dB PSNR gain. Our strategy also outperforms DIP with at least a + 1.52 dB PSNR advantage and peaks at + 3.15 dB with fewer angles. Moreover, cGLO does not create artifacts or structural deformations contrary to DIP and SGMs.

CONCLUSIONS: We propose a parsimonious and robust reconstruction technique offering similar to better performances when compared with state-of-the-art methods regarding full-dose sparse-view CT. Our strategy could be readily applied to any imaging reconstruction task without any assumption about the acquisition protocol or the quantity of available data.

PMID:40177097 | PMC:PMC11961077 | DOI:10.1117/1.JMI.12.2.024004

Front Artif Intell. 2025 Mar 18;8:1565287. doi: 10.3389/frai.2025.1565287. eCollection 2025.

ABSTRACT

Vehicle-to-Everything (V2X) communication promises to revolutionize road safety and efficiency. However, challenges in data sharing and network reliability impede its full realization. This paper addresses these challenges by proposing a novel Deep Learning (DL) approach for traffic prediction in V2X environments. We employ Bidirectional Long Short-Term Memory (BiLSTM) networks and compare their performance against other prominent DL architectures, including unidirectional LSTM and Gated Recurrent Unit (GRU). Our findings demonstrate that the BiLSTM model exhibits superior accuracy in predicting traffic patterns. This enhanced prediction capability enables more efficient resource allocation, improved network performance, and enhanced safety for all road users, reducing fuel consumption, decreased emissions, and a more sustainable transportation system.

PMID:40176965 | PMC:PMC11962783 | DOI:10.3389/frai.2025.1565287

J Sleep Res. 2025 Apr 3:e70061. doi: 10.1111/jsr.70061. Online ahead of print.

ABSTRACT

Automated sleep staging on wearable data could improve our understanding and management of epilepsy. This study evaluated sleep scoring by a deep learning model on 223 night-sleep recordings from 50 patients measured in the hospital with an electroencephalogram (EEG) and a wearable device. The model scored the sleep stage of every 30-s epoch on the EEG and wearable data, and we compared the output with a clinical expert on 20 nights, each for a different patient. The Bland-Altman analysis examined differences in the automated staging in both modalities, and using mixed-effect models, we explored sleep differences between patients with and without seizures. Overall, we found moderate accuracy and Cohen's kappa on the model scoring of standard EEG (0.73 and 0.59) and the wearable (0.61 and 0.43) versus the clinical expert. F1 scores also varied between patients and the modalities. The sensitivity varied by sleep stage and was very low for stage N1. Moreover, sleep staging on the wearable data underestimated the duration of most sleep macrostructure parameters except N2. On the other hand, patients with seizures during the hospital admission slept more each night (6.37, 95% confidence interval [CI] 5.86-7.87) compared with patients without seizures (5.68, 95% CI 5.24-6.13), p = 0.001, but also spent more time in stage N2. In conclusion, wearable EEG and accelerometry could monitor sleep in patients with epilepsy, and our approach can help automate the analysis. However, further steps are essential to improve the model performance before clinical implementation. Trial Registration: The SeizeIT2 trial was registered in clinicaltrials.gov, NCT04284072.

PMID:40176726 | DOI:10.1111/jsr.70061

Sci Rep. 2025 Apr 2;15(1):11236. doi: 10.1038/s41598-025-89911-3.

ABSTRACT

The early detection of brain tumors is very important for treating them and improving the quality of life for patients. Through advanced imaging techniques, doctors can now make more informed decisions. This paper introduces a framework for a tumor detection system capable of grading gliomas. The system's implementation begins with the acquisition and analysis of brain magnetic resonance images. Key features indicative of tumors and gliomas are extracted and classified as independent components. A deep learning model is then employed to categorize these gliomas. The proposed model classifies gliomas into three primary categories: meningioma, pituitary, and glioma. Performance evaluation demonstrates a high level of accuracy (99.21%), specificity (98.3%), and sensitivity (97.83%). Further research and validation are essential to refine the system and ensure its clinical applicability. The development of accurate and efficient tumor detection systems holds significant promise for enhancing patient care and improving survival rates.

PMID:40175410 | DOI:10.1038/s41598-025-89911-3

Rev Neurol (Paris). 2025 Apr 1:S0035-3787(25)00487-4. doi: 10.1016/j.neurol.2025.03.006. Online ahead of print.

ABSTRACT

In recent years, artificial intelligence (AI) has become an increasingly prominent focus of medical research, significantly impacting epileptology as well. Studies on deep learning (DL) and machine learning (ML) - the core of AI - have explored their applications in epilepsy imaging, primarily focusing on lesion detection, lateralization and localization of epileptogenic areas, postsurgical outcome prediction and automatic differentiation between people with epilepsy and healthy individuals. Various AI-driven approaches are being investigated across different neuroimaging modalities, with the ultimate goal of integrating these tools into clinical practice to enhance the diagnosis and treatment of epilepsy. As computing power continues to advance, the development, research integration, and clinical implementation of AI applications are expected to accelerate, making them even more effective and accessible. However, ensuring the safety of patient data will require strict regulatory measures. Despite these challenges, AI represents a transformative opportunity for medicine, particularly in epilepsy neuroimaging. Since ML and DL models thrive on large datasets, fostering collaborations and expanding open-access databases will become increasingly pivotal in the future.

PMID:40175210 | DOI:10.1016/j.neurol.2025.03.006

Acad Radiol. 2025 Apr 1:S1076-6332(25)00210-7. doi: 10.1016/j.acra.2025.03.015. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: Routine knee MRI protocols for 1.5 T and 3 T scanners, do not include T2*-w gradient echo (T2*W) images, which are useful in several clinical scenarios such as the assessment of cartilage, synovial blooming (deposition of hemosiderin), chondrocalcinosis and the evaluation of the physis in pediatric patients. Herein, we aimed to develop an open-source deep learning model that creates synthetic T2*W images of the knee using fat-suppressed intermediate-weighted images.

MATERIALS AND METHODS: A cycleGAN model was trained with 12,118 sagittal knee MR images and tested on an independent set of 2996 images. Diagnostic interchangeability of synthetic T2*W images was assessed against a series of findings. Voxel intensity of four tissues was evaluated with Bland-Altman plots. Image quality was assessed with the use of root mean squared error (NRMSE), structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR). Code, model and a standalone executable file are provided on github.

RESULTS: The model achieved a median NRMSE, PSNR and SSIM of 0.5, 17.4, and 0.5, respectively. Images were found interchangeable with an intraclass correlation coefficient >0.95 for all findings. Mean voxel intensity was equal between synthetic and conventional images. Four types of artifacts were identified: geometrical distortion (86/163 cases), object insertion/omission (11/163 cases), a wrap-around-like (26/163 cases) and an incomplete fat-suppression artifact (120/163 cases), which had a median 0 impact (no impact) on the diagnosis.

CONCLUSION: In conclusion, the developed open-source GAN model creates synthetic T2*W images of the knee of high diagnostic value and quality. The identified artifacts had no or minor effect on the diagnostic value of the images.

PMID:40175204 | DOI:10.1016/j.acra.2025.03.015

Adv Pharmacol. 2025;103:325-348. doi: 10.1016/bs.apha.2025.01.016. Epub 2025 Feb 16.

ABSTRACT

Artificial Intelligence (AI) has revolutionized drug discovery by enhancing data collection, integration, and predictive modeling across various critical stages. It aggregates vast biological and chemical data, including genomic information, protein structures, and chemical interactions with biological targets. Machine learning techniques and QSAR models are applied by AI to predict compound behaviors and predict potential drug candidates. Docking simulations predict drug-protein interactions, while virtual screening eliminates large chemical databases through efficient sifting. Similarly, AI supports de novo drug design by generating novel molecules, optimized against a particular biological target, using generative models such as generative adversarial network (GAN), always finding lead compounds with the most desirable pharmacological properties. AI used in clinical trials improves efficiency by pinpointing responsive patient cohorts leveraging genetic profiles and biomarkers and maintaining propriety such as dataset diversity and compliance with regulations. This chapter aimed to summarize and analyze the mechanism of AI to accelerate drug discovery by streamlining different processes that enable informed decisions and bring potential life-saving therapies to market faster, amounting to a breakthrough in pharmaceutical research and development.

PMID:40175048 | DOI:10.1016/bs.apha.2025.01.016

Adv Pharmacol. 2025;103:101-120. doi: 10.1016/bs.apha.2025.01.008. Epub 2025 Feb 6.

ABSTRACT

The lengthy and costly drug discovery process is transformed by deep learning, a subfield of artificial intelligence. Deep learning technologies expedite the procedure, increasing treatment success rates and speeding life-saving procedures. Deep learning stands out in target identification and lead selection. Deep learning greatly accelerates initial stage by analyzing large datasets of biological data to identify possible therapeutic targets and rank targeted drug molecules with desired features. Predicting possible adverse effects is another significant challenge. Deep learning offers prompt and efficient assistance with toxicology prediction in a very short time, deep learning algorithms can forecast a new drug's possible harm. This enables to concentrate on safer alternatives and steer clear of late-stage failures brought on by unanticipated toxicity. Deep learning unlocks the possibility of drug repurposing; by examining currently available medications, it is possible to find whole new therapeutic uses. This method speeds up development of diseases that were previously incurable. De novo drug discovery is made possible by deep learning when combined with sophisticated computational modeling, it can create completely new medications from the ground. Deep learning can recommend and direct towards new drug candidates with high binding affinities and intended therapeutic effects by examining molecular structures of disease targets. This provides focused and personalized medication. Lastly, drug characteristics can be optimized with aid of deep learning. Researchers can create medications with higher bioavailability and fewer toxicity by forecasting drug pharmacokinetics. In conclusion, deep learning promises to accelerate drug development, reduce costs, and ultimately save lives.

PMID:40175037 | DOI:10.1016/bs.apha.2025.01.008

Alzheimers Res Ther. 2025 Apr 2;17(1):70. doi: 10.1186/s13195-025-01705-x.

ABSTRACT

BACKGROUND: Although interactions between amyloid-beta and tau proteins have been implicated in Alzheimer's disease (AD), the precise mechanisms by which these interactions contribute to disease progression are not yet fully understood. Moreover, despite the growing application of deep learning in various biomedical fields, its application in integrating networks to analyze disease mechanisms in AD research remains limited. In this study, we employed BIONIC, a deep learning-based network integration method, to integrate proteomics and protein-protein interaction data, with an aim to uncover factors that moderate the effects of the Aβ-tau interaction on mild cognitive impairment (MCI) and early-stage AD.

METHODS: Proteomic data from the ROSMAP cohort were integrated with protein-protein interaction (PPI) data using a Deep Learning-based model. Linear regression analysis was applied to histopathological and gene expression data, and mutual information was used to detect moderating factors. Statistical significance was determined using the Benjamini-Hochberg correction (p < 0.05).

RESULTS: Our results suggested that astrocytes and GPNMB + microglia moderate the Aβ-tau interaction. Based on linear regression with histopathological and gene expression data, GFAP and IBA1 levels and GPNMB gene expression positively contributed to the interaction of tau with Aβ in non-dementia cases, replicating the results of the network analysis.

CONCLUSIONS: These findings suggest that GPNMB + microglia moderate the Aβ-tau interaction in early AD and therefore are a novel therapeutic target. To facilitate further research, we have made the integrated network available as a visualization tool for the scientific community (URL: https://igcore.cloud/GerOmics/AlzPPMap ).

PMID:40176187 | DOI:10.1186/s13195-025-01705-x

Cancer Imaging. 2025 Apr 2;25(1):47. doi: 10.1186/s40644-025-00869-x.

ABSTRACT

OBJECTIVE: This study evaluates the impact of deep learning-enhanced T1-weighted VIBE sequences (DL-VIBE) on image quality and procedural parameters during MR-guided thermoablation of liver malignancies, compared to standard VIBE (SD-VIBE).

METHODS: Between September 2021 and February 2023, 34 patients (mean age: 65.4 years; 13 women) underwent MR-guided microwave ablation on a 1.5 T scanner. Intraprocedural SD-VIBE sequences were retrospectively processed with a deep learning algorithm (DL-VIBE) to reduce noise and enhance sharpness. Two interventional radiologists independently assessed image quality, noise, artifacts, sharpness, diagnostic confidence, and procedural parameters using a 5-point Likert scale. Interrater agreement was analyzed, and noise maps were created to assess signal-to-noise ratio improvements.

RESULTS: DL-VIBE significantly improved image quality, reduced artifacts and noise, and enhanced sharpness of liver contours and portal vein branches compared to SD-VIBE (p < 0.01). Procedural metrics, including needle tip detectability, confidence in needle positioning, and ablation zone assessment, were significantly better with DL-VIBE (p < 0.01). Interrater agreement was high (Cohen κ = 0.86). Reconstruction times for DL-VIBE were 3 s for k-space reconstruction and 1 s for superresolution processing. Simulated acquisition modifications reduced breath-hold duration by approximately 2 s.

CONCLUSION: DL-VIBE enhances image quality during MR-guided thermal ablation while improving efficiency through reduced processing and acquisition times.

PMID:40176185 | DOI:10.1186/s40644-025-00869-x

Plant Methods. 2025 Apr 2;21(1):48. doi: 10.1186/s13007-025-01325-4.

ABSTRACT

Plant leaf diseases significantly threaten agricultural productivity and global food security, emphasizing the importance of early and accurate detection and effective crop health management. Current deep learning models, often used for plant disease classification, have limitations in capturing intricate features such as texture, shape, and color of plant leaves. Furthermore, many of these models are computationally expensive and less suitable for deployment in resource-constrained environments such as farms and rural areas. We propose a novel Lightweight Deep Learning model, Depthwise Separable Convolution with Spatial Attention (LWDSC-SA), designed to address limitations and enhance feature extraction while maintaining computational efficiency. By integrating spatial attention and depthwise separable convolution, the LWDSC-SA model improves the ability to detect and classify plant diseases. In our comprehensive evaluation using the PlantVillage dataset, which consists of 38 classes and 55,000 images from 14 plant species, the LWDSC-SA model achieved 98.7% accuracy. It presents a substantial improvement over MobileNet by 5.25%, MobileNetV2 by 4.50%, AlexNet by 7.40%, and VGGNet16 by 5.95%. Furthermore, to validate its robustness and generalizability, we employed K-fold cross-validation K=5, which demonstrated consistently high performance, with an average accuracy of 98.58%, precision of 98.30%, recall of 98.90%, and F1 score of 98.58%. These results highlight the superior performance of the proposed model, demonstrating its ability to outperform state-of-the-art models in terms of accuracy while remaining lightweight and efficient. This research offers a promising solution for real-world agricultural applications, enabling effective plant disease detection in resource-limited settings and contributing to more sustainable agricultural practices.

PMID:40176127 | DOI:10.1186/s13007-025-01325-4

Biomed Eng Online. 2025 Apr 2;24(1):39. doi: 10.1186/s12938-025-01360-1.

ABSTRACT

BACKGROUND: Hands-and-knees crawling is a promising rehabilitation intervention for infants with motor impairments, while research on assistive crawling devices for rehabilitation training was still in its early stages. In particular, precisely generating motion trajectories is a prerequisite to controlling exoskeleton assistive devices, and deep learning-based prediction algorithms, such as Long-Short-Term Memory (LSTM) networks, have proven effective in forecasting joint trajectories of gait. Despite this, no previous studies have focused on forecasting the more variable and complex trajectories of infant crawling. Therefore, this paper aims to explore the feasibility of using LSTM networks to predict crawling trajectories, thereby advancing our understanding of how to actively control crawling rehabilitation training robots.

METHODS: We collected joint trajectory data from 20 healthy infants (11 males and 9 females, aged 8-15 months) as they crawled on hands and knees. This study implemented LSTM networks to forecast bilateral elbow and knee trajectories based on corresponding joint angles. The data set comprised 58, 782 time steps, each containing 4 joint angles. We partitioned the data set into 70% for training and 30% for testing to evaluate predictive performance. We investigated a total of 24 combinations of input and output time-frames, with window sizes for input vectors ranging from 10, 15, 20, 30, 40, 50, 70, and 100 time steps, and output vectors from 5, 10, and 15 steps. Evaluation metrics included Mean Absolute Error (MAE), Mean Squared Error (MSE), and Correlation Coefficient (CC) to assess prediction accuracy.

RESULTS: The results indicate that across various input-output windows, the MAE for elbow joints ranged from 0.280 to 4.976°, MSE ranged from 0.203° to 59.186°, and CC ranged from 89.977% to 99.959%. For knee joints, MAE ranged from 0.277 to 4.262°, MSE from 0.229 to 53.272°, and CC from 89.454% to 99.944%. Results also show that smaller output window sizes lead to lower prediction errors. As expected, the LSTM predicting 5 output time steps has the lowest average error, while the LSTM predicting 15 time steps has the highest average error. In addition, variations in input window size had a minimal impact on average error when the output window size was fixed. Overall, the optimal performance for both elbow and knee joints was observed with input-output window sizes of 30 and 5 time steps, respectively, yielding an MAE of 0.295°, MSE of 0.260°, and CC of 99.938%.

CONCLUSIONS: This study demonstrates the feasibility of forecasting infant crawling trajectories using LSTM networks, which could potentially integrate with exoskeleton control systems. It experimentally explores how different input and output time-frames affect prediction accuracy and sets the stage for future research focused on optimizing models and developing effective control strategies to improve assistive crawling devices.

PMID:40176123 | DOI:10.1186/s12938-025-01360-1

J Imaging Inform Med. 2025 Apr 2. doi: 10.1007/s10278-025-01489-4. Online ahead of print.

ABSTRACT

Chronic kidney disease (CKD) remains a major public health concern, requiring better predictive models for early intervention. This study evaluates a deep learning model (DLM) that utilizes raw chest X-ray (CXR) data to predict moderate to severe kidney function decline. We analyzed data from 79,219 patients with an estimated Glomerular Filtration Rate (eGFR) between 65 and 120, segmented into development (n = 37,983), tuning (n = 15,346), internal validation (n = 14,113), and external validation (n = 11,777) sets. Our DLM, pretrained on CXR-report pairs, was fine-tuned with the development set. We retrospectively examined data spanning April 2011 to February 2022, with a 5-year maximum follow-up. Primary and secondary endpoints included CKD stage 3b progression, ESRD/dialysis, and mortality. The overall concordance index (C-index) values for the internal and external validation sets were 0.903 (95% CI, 0.885-0.922) and 0.851 (95% CI, 0.819-0.883), respectively. In these sets, the incidences of progression to CKD stage 3b at 5 years were 19.2% and 13.4% in the high-risk group, significantly higher than those in the median-risk (5.9% and 5.1%) and low-risk groups (0.9% and 0.9%), respectively. The sex, age, and eGFR-adjusted hazard ratios (HR) for the high-risk group compared to the low-risk group were 16.88 (95% CI, 10.84-26.28) and 7.77 (95% CI, 4.77-12.64), respectively. The high-risk group also exhibited higher probabilities of progressing to ESRD/dialysis or experiencing mortality compared to the low-risk group. Further analysis revealed that the high-risk group compared to the low/median-risk group had a higher prevalence of complications and abnormal blood/urine markers. Our findings demonstrate that a DLM utilizing CXR can effectively predict CKD stage 3b progression, offering a potential tool for early intervention in high-risk populations.

PMID:40175823 | DOI:10.1007/s10278-025-01489-4

Microb Ecol. 2025 Apr 2;88(1):22. doi: 10.1007/s00248-025-02516-1.

ABSTRACT

Shifts in agricultural land use over the past 200 years have led to a loss of nearly 50% of existing wetlands in the USA, and agricultural activities contribute up to 65% of the nutrients that reach the Mississippi River Basin, directly contributing to biological disasters such as the hypoxic Gulf of Mexico "Dead" Zone. Federal efforts to construct and restore wetland habitats have been employed to mitigate the detrimental effects of eutrophication, with an emphasis on the restoration of ecosystem services such as nutrient cycling and retention. Soil microbial assemblages drive biogeochemical cycles and offer a unique and sensitive framework for the accurate evaluation, restoration, and management of ecosystem services. The purpose of this study was to elucidate patterns of soil bacteria within and among wetlands by developing diversity profiles from high-throughput sequencing data, link functional gene copy number of nitrogen cycling genes to measured nutrient flux rates collected from flow-through incubation cores, and predict nutrient flux using microbial assemblage composition. Soil microbial assemblages showed fine-scale turnover in soil cores collected across the topsoil horizon (0-5 cm; top vs bottom partitions) and were structured by restoration practices on the easements (tree planting, shallow water, remnant forest). Connections between soil assemblage composition, functional gene copy number, and nutrient flux rates show the potential for soil bacterial assemblages to be used as bioindicators for nutrient cycling on the landscape. In addition, the predictive accuracy of flux rates was improved when implementing deep learning models that paired connected samples across time.

PMID:40175811 | DOI:10.1007/s00248-025-02516-1

Nat Rev Cancer. 2025 Apr 2. doi: 10.1038/s41568-025-00811-0. Online ahead of print.

NO ABSTRACT

PMID:40175619 | DOI:10.1038/s41568-025-00811-0