Int J Lab Hematol. 2025 Mar 2. doi: 10.1111/ijlh.14455. Online ahead of print.

ABSTRACT

INTRODUCTION: Von Willebrand factor (VWF) multimer analysis is essential for diagnosing and classifying von Willebrand disease (VWD) but requires expert interpretation and is subject to inter-rater variability. We developed an automated image analysis pipeline using deep learning to improve the reproducibility and efficiency of VWF multimer pattern classification.

METHODS: We trained a YOLOv8 deep learning model on 514 gel images (6168 labeled instances) to classify VWF multimer patterns into 12 classes. The model was validated on 192 images (2304 instances) and tested on an independent set of 94 images (1128 instances). Images underwent preprocessing, including histogram equalization, contrast enhancement, and gamma correction. Two expert raters provided ground truth classifications.

RESULTS: The model achieved 91% accuracy compared to Expert 1 (macro-averaged precision = 0.851, recall = 0.757, F1-score = 0.786) and 87% accuracy compared to Expert 2 (macro-averaged precision = 0.653, recall = 0.653, F1-score = 0.641). Inter-rater agreement was very high between experts (κ = 0.883), with strong agreement between the model and Expert 1 (κ = 0.845) and good agreement with Expert 2 (κ = 0.773). The model performed exceptionally well on common patterns (F1 > 0.93) but showed lower performance on rare subtypes.

CONCLUSION: Automated VWF multimer analysis using deep learning demonstrates high accuracy in pattern classification and could standardize the interpretation of VWF multimer patterns. While not replacing expert analysis, this approach could improve the efficiency of expert human review, potentially streamlining laboratory workflow and expanding access to VWF multimer testing.

PMID:40025642 | DOI:10.1111/ijlh.14455

Sci Rep. 2025 Mar 1;15(1):7266. doi: 10.1038/s41598-025-91786-3.

ABSTRACT

Cervical cancer, arising from the cells of the cervix, the lower segment of the uterus connected to the vagina-poses a significant health threat. The microscopic examination of cervical cells using Pap smear techniques plays a crucial role in identifying potential cancerous alterations. While developed nations demonstrate commendable efficiency in Pap smear acquisition, the process remains laborious and time-intensive. Conversely, in less developed regions, there is a pressing need for streamlined, computer-aided methodologies for the pre-analysis and treatment of cervical cancer. This study focuses on the classification of squamous cells into five distinct classes, providing a nuanced assessment of cervical cancer severity. Utilizing a dataset comprising over 4096 images from SimpakMed, available on Kaggle, we employed ensemble technique which included the Convolutional Neural Network (CNN), AlexNet, and SqueezeNet for image classification, achieving accuracies of 90.8%, 92%, and 91% respectively. Particularly noteworthy is the proposed ensemble technique, which surpasses individual model performances, achieving an impressive accuracy of 94%. This ensemble approach underscores the efficacy of our method in precise squamous cell classification and, consequently, in gauging the severity of cervical cancer. The results represent a promising advancement in the development of more efficient diagnostic tools for cervical cancer in resource-constrained settings.

PMID:40025091 | DOI:10.1038/s41598-025-91786-3

Sci Rep. 2025 Mar 1;15(1):7291. doi: 10.1038/s41598-025-89612-x.

ABSTRACT

The traditional molecular-based identification (TMID) technique of new infections from genome sequences (GSs) has made significant contributions so far. However, due to the sensitive nature of the medical data, the TMID technique of transferring the patient's data to the central machine or server may create severe privacy and security issues. In recent years, the progression of deep federated learning (DFL) and its remarkable success in many domains has guided as a potential solution in this field. Therefore, we proposed a dependable and privacy-preserving DFL-based identification model of new infections from GSs. The unique contributions include automatic effective feature selection, which is best suited for identifying new infections, designing a dependable and privacy-preserving DFL-based LeNet model, and evaluating real-world data. To this end, a comprehensive experimental performance evaluation has been conducted. Our proposed model has an overall accuracy of 99.12% after independently and identically distributing the dataset among six clients. Moreover, the proposed model has a precision of 98.23%, recall of 98.04%, f1-score of 96.24%, Cohen's kappa of 83.94%, and ROC AUC of 98.24% for the same configuration, which is a noticeable improvement when compared to the other benchmark models. The proposed dependable model, along with empirical results, is encouraging enough to recognize as an alternative for identifying new infections from other virus strains by ensuring proper privacy and security of patients' data.

PMID:40025035 | DOI:10.1038/s41598-025-89612-x

Biomed Eng Online. 2025 Mar 2;24(1):27. doi: 10.1186/s12938-025-01355-y.

ABSTRACT

BACKGROUND: To develop and validate a model that integrates clinical data, deep learning radiomics, and radiomic features to predict high-risk patients for cage subsidence (CS) after lumbar fusion.

METHODS: This study analyzed preoperative CT and MRI data from 305 patients undergoing lumbar fusion surgery from three centers. Using a deep learning model based on 3D vision transformations, the data were divided the dataset into training (n = 214), validation (n = 61), and test (n = 30) groups. Feature selection was performed using LASSO regression, followed by the development of a logistic regression model. The predictive ability of the model was assessed using various machine learning algorithms, and a combined clinical model was also established.

RESULTS: Ultimately, 11 traditional radiomic features, 5 deep learning radiomic features, and 1 clinical feature were selected. The combined model demonstrated strong predictive performance, with area under the curve (AUC) values of 0.941, 0.832, and 0.935 for the training, validation, and test groups, respectively. Notably, our model outperformed predictions made by two experienced surgeons.

CONCLUSIONS: This study developed a robust predictive model that integrates clinical features and imaging data to identify high-risk patients for CS following lumbar fusion. This model has the potential to improve clinical decision-making and reduce the need for revision surgeries, easing the burden on healthcare systems.

PMID:40025592 | DOI:10.1186/s12938-025-01355-y

J Cheminform. 2025 Mar 2;17(1):27. doi: 10.1186/s13321-025-00975-9.

ABSTRACT

Molecular optimization is a crucial step in drug development, involving structural modifications to improve the desired properties of drug candidates. Although many deep-learning-based molecular optimization algorithms have been proposed and may perform well on benchmarks, they usually do not pay sufficient attention to the synthesizability of molecules, resulting in optimized compounds difficult to be synthesized. To address this issue, we first developed a general pipeline capable of constructing functional reaction template library specific to any property where a predictive model can be built. Based on these functional templates, we introduced Syn-MolOpt, a synthesis planning-oriented molecular optimization method. During optimization, functional reaction templates steer the process towards specific properties by effectively transforming relevant structural fragments. In four diverse tasks, including two toxicity-related (GSK3β-Mutagenicity and GSK3β-hERG) and two metabolism-related (GSK3β-CYP3A4 and GSK3β-CYP2C19) multi-property molecular optimizations, Syn-MolOpt outperformed three benchmark models (Modof, HierG2G, and SynNet), highlighting its efficacy and adaptability. Additionally, visualization of the synthetic routes for molecules optimized by Syn-MolOpt confirms the effectiveness of functional reaction templates in molecular optimization. Notably, Syn-MolOpt's robust performance in scenarios with limited scoring accuracy demonstrates its potential for real-world molecular optimization applications. By considering both optimization and synthesizability, Syn-MolOpt promises to be a valuable tool in molecular optimization.Scientific contribution Syn-MolOpt takes into account both molecular optimization and synthesis, allowing for the design of property-specific functional reaction template libraries for the properties to be optimized, and providing reference synthesis routes for the optimized compounds while optimizing the targeted properties. Syn-MolOpt's universal workflow makes it suitable for various types of molecular optimization tasks.

PMID:40025591 | DOI:10.1186/s13321-025-00975-9

J Cheminform. 2025 Mar 2;17(1):28. doi: 10.1186/s13321-025-00973-x.

ABSTRACT

Computer-aided drug design has the potential to significantly reduce the astronomical costs of drug development, and molecular docking plays a prominent role in this process. Molecular docking is an in silico technique that predicts the bound 3D conformations of two molecules, a necessary step for other structure-based methods. Here, we describe version 1.3 of the open-source molecular docking software GNINA. This release updates the underlying deep learning framework to PyTorch, resulting in more computationally efficient docking and paving the way for seamless integration of other deep learning methods into the docking pipeline. We retrained our CNN scoring functions on the updated CrossDocked2020 v1.3 dataset and introduce knowledge-distilled CNN scoring functions to facilitate high-throughput virtual screening with GNINA. Furthermore, we add functionality for covalent docking, where an atom of the ligand is covalently bound to an atom of the receptor. This update expands the scope of docking with GNINA and further positions GNINA as a user-friendly, open-source molecular docking framework. GNINA is available at https://github.com/gnina/gnina .Scientific contributions: GNINA 1.3 is an open source a molecular docking tool with enhanced support for covalent docking and updated deep learning models for more effective docking and screening.

PMID:40025560 | DOI:10.1186/s13321-025-00973-x

BMC Oral Health. 2025 Mar 1;25(1):329. doi: 10.1186/s12903-025-05677-0.

ABSTRACT

BACKGROUND: Several commercial programs incorporate artificial intelligence in diagnosis, but very few dental professionals have been surveyed regarding its acceptability and usability. Furthermore, few have explored how these advances might be incorporated into routine practice.

METHODS: Our team developed and implemented a deep learning (DL) model employing semantic segmentation neural networks and object detection networks to precisely identify alveolar bone crestal levels (ABCLs) and cemento-enamel junctions (CEJs) to measure change in alveolar crestal height (ACH). The model was trained and validated using a 550 bitewing radiograph dataset curated by an oral radiologist, setting a gold standard for ACH measurements. A twenty-question survey was created to compare the accuracy and efficiency of manual X-ray examination versus the application and to assess the acceptability and usability of the application.

RESULTS: In total, 56 different dental professionals classified severe (ACH > 5 mm) vs. non-severe (ACH ≤ 5 mm) periodontal bone loss on 35 calculable ACH measures. Dental professionals accurately identified between 35-87% of teeth with severe periodontal disease, whereas the artificial intelligence (AI) application achieved an 82-87% accuracy rate. Among the 65 participants who completed the acceptability and usability survey, more than half the participants (52%) were from an academic setting. Only 21% of participants reported that they already used automated or AI-based software in their practice to assist in reading of X-rays. The majority, 57%, stated that they only approximate when measuring bone levels and only 9% stated that they measure with a ruler. The survey indicated that 84% of participants agreed or strongly agreed with the AI application measurement of ACH. Furthermore, 56% of participants agreed that AI would be helpful in their professional setting.

CONCLUSION: Overall, the study demonstrates that an AI application for detecting alveolar bone has high acceptability among dental professionals and may provide benefits in time saving and increased clinical accuracy.

PMID:40025477 | DOI:10.1186/s12903-025-05677-0

BMC Oral Health. 2025 Mar 1;25(1):328. doi: 10.1186/s12903-024-05231-4.

ABSTRACT

BACKGROUND: The aim of our study was to develop and evaluate a deep learning model (BiStageNet) for automatic detection of dens evaginatus (DE) premolars on orthodontic intraoral photographs. Additionally, based on the training results, we developed a DE detection platform for orthodontic clinical applications.

METHODS: We manually selected the premolar areas for automatic premolar recognition training using a dataset of 1,400 high-quality intraoral photographs. Next, we labeled each premolar for DE detection training using a dataset of 2,128 images. We introduced the Dice coefficient, accuracy, sensitivity, specificity, F1-score, ROC curve as well as areas under the ROC curve to evaluate the learning results of our model. Finally, we constructed an automatic DE detection platform based on our trained model (BiStageNet) using Pytorch.

RESULTS: Our DE detection platform achieved a mean Dice coefficient of 0.961 in premolar recognition, with a diagnostic accuracy of 85.0%, sensitivity of 88.0%, specificity of 82.0%, F1 Score of 0.854, and AUC of 0.93. Experimental results revealed that dental interns, when manually identifying DE, showed low specificity. With the tool's assistance, specificity significantly improved for all interns, effectively reducing false positives without sacrificing sensitivity. This led to enhanced diagnostic precision, evidenced by improved PPV, NPV, and F1-Scores.

CONCLUSION: Our BiStageNet was capable of recognizing premolars and detecting DE with high accuracy on intraoral photographs. On top of that, our self-developed DE detection platform was promising for clinical application and promotion.

PMID:40025464 | DOI:10.1186/s12903-024-05231-4

Commun Med (Lond). 2025 Mar 1;5(1):55. doi: 10.1038/s43856-025-00767-0.

ABSTRACT

BACKGROUND: Glioblastoma is a highly heterogeneous brain tumor, posing challenges for precision therapies and patient stratification in clinical trials. Understanding how genetic mutations influence tumor imaging may improve patient management and treatment outcomes. This study investigates the relationship between imaging features, spatial patterns of tumor location, and genetic alterations in IDH-wildtype glioblastoma, as well as the likely sequence of mutational events.

METHODS: We conducted a retrospective analysis of 357 IDH-wildtype glioblastomas with pre-operative multiparametric MRI and targeted genetic sequencing data. Radiogenomic signatures and spatial distribution maps were generated for key mutations in genes such as EGFR, PTEN, TP53, and NF1 and their corresponding pathways. Machine and deep learning models were used to identify imaging biomarkers and stratify tumors based on their genetic profiles and molecular heterogeneity.

RESULTS: Here, we show that glioblastoma mutations produce distinctive imaging signatures, which are more pronounced in tumors with less molecular heterogeneity. These signatures provide insights into how mutations affect tumor characteristics such as neovascularization, cell density, invasion, and vascular leakage. We also found that tumor location and spatial distribution correlate with genetic profiles, revealing associations between tumor regions and specific oncogenic drivers. Additionally, imaging features reflect the cross-sectionally inferred evolutionary trajectories of glioblastomas.

CONCLUSIONS: This study establishes clinically accessible imaging biomarkers that capture the molecular composition and oncogenic drivers of glioblastoma. These findings have potential implications for noninvasive tumor profiling, personalized therapies, and improved patient stratification in clinical trials.

PMID:40025245 | DOI:10.1038/s43856-025-00767-0

Sci Rep. 2025 Mar 1;15(1):7301. doi: 10.1038/s41598-025-91823-1.

ABSTRACT

The classification of tongue shapes is essential for objective tongue diagnoses. However, the accuracy of classification is influenced by numerous factors. First, considerable differences exist between individuals with the same tongue shape. Second, the lips interfere with tongue shape classification. Additionally, small datasets make it difficult to conduct network training. To address these issues, this study builds a two-level nested tongue segmentation and tongue image classification network named IF-RCNet based on feature fusion and mixed input methods. In IF-RCNet, RCA-UNet is used to segment the tongue body, and RCA-Net is used to classify the tongue shape. The feature fusion strategy can enhance the network's ability to extract tongue features, and the mixed input can expand the data input of RCA-Net. The experimental results show that tongue shape classification based on IF-RCNet outperforms many other classification networks (VGG 16, ResNet 18, AlexNet, ViT and MobileNetv4). The method can accurately classify tongues despite the negative effects of differences between homogeneous tongue shapes and the misclassification of normal versus bulgy tongues due to lip interference. The method exhibited better performance on a small dataset of tongues, thereby enhancing the accuracy of tongue shape classification and providing a new approach for tongue shape classification.

PMID:40025207 | DOI:10.1038/s41598-025-91823-1

Sci Rep. 2025 Mar 1;15(1):7337. doi: 10.1038/s41598-025-92105-6.

ABSTRACT

Machine learning approaches including deep learning models have shown promising performance in the automatic detection of Parkinson's disease. These approaches rely on different types of data with voice recordings being the most used due to the convenient and non-invasive nature of data acquisition. Our group has successfully developed a novel approach that uses convolutional neural network with transfer learning to analyze spectrogram images of the sustained vowel /a/ to identify people with Parkinson's disease. We tested this approach by collecting a dataset of voice recordings via analog telephone lines, which support limited bandwidth. The convolutional neural network with transfer learning approach showed superior performance against conventional machine learning methods that collapse measurements across time to generate feature vectors. This study builds upon our prior results and presents two novel contributions: First, we tested the performance of our approach on a larger voice dataset recorded using smartphones with wide bandwidth. Our results show comparable performance between two datasets generated using different recording platforms despite the differences in most important features resulting from the limited bandwidth of analog telephonic lines. Second, we compared the classification performance achieved using linear-scale and mel-scale spectrogram images and showed a small but statistically significant gain using mel-scale spectrograms.

PMID:40025201 | DOI:10.1038/s41598-025-92105-6

NPJ Digit Med. 2025 Mar 1;8(1):133. doi: 10.1038/s41746-025-01527-z.

ABSTRACT

This systematic review evaluated natural language processing (NLP) approaches for detecting cognitive impairment in electronic health record clinical notes. Following PRISMA guidelines, we analyzed 18 studies (n = 1,064,530) that employed rule-based algorithms (67%), traditional machine learning (28%), and deep learning (17%). NLP models demonstrated robust performance in identifying cognitive decline, with median sensitivity 0.88 (IQR 0.74-0.91) and specificity 0.96 (IQR 0.81-0.99). Deep learning architectures achieved superior results, with area under the receiver operating characteristic curves up to 0.997. Major implementation challenges included incomplete electronic health record data capture, inconsistent clinical documentation practices, and limited external validation. While NLP demonstrates promise, successful clinical translation requires establishing standardized approaches, improving access to annotated datasets, and developing equitable deployment frameworks.

PMID:40025194 | DOI:10.1038/s41746-025-01527-z

Sci Rep. 2025 Mar 1;15(1):7327. doi: 10.1038/s41598-025-91839-7.

ABSTRACT

As the water economy advances and the concepts of water ecology protection and sustainable development take root in people's minds, underwater imaging equipment has made remarkable progress. However, due to various factors, underwater images still suffer from low quality. How to enhance the quality of underwater images so that people can understand them quickly has become a crucial issue. Therefore, aiming at the degradation problems such as detail blurring, color imbalance, and noise interference in low-quality underwater images, this paper proposes an optimized UNet framework with a joint loss function (OUNet-JL). Firstly, to alleviate the problem of detail blurring, we construct a multi-residual module (MRM) to enhance the ability to represent detail features by using serially stacked convolutional blocks and residual connections. Secondly, we build a spatial multi-scale feature extraction module fused with channel attention (SMFM) to address the color imbalance issue through multi-scale dilated convolution and channel attention. Thirdly, to improve the signal-to-noise ratio of the enhanced image and solve the problem of blurring distortion, a strengthen-operate-subtract feature reconstruction module (SOSFM) is presented. Fourthly, to guide the network to perform training more efficiently and help it converge rapidly, a joint loss function is designed by integrating four different loss functions. Extensive experiments conducted on the well-known UIEB and UFO-120 datasets have shown the superiority of our OUNet-JL compared with several state-of-the-art algorithms. Moreover, ablation studies have also verified the effectiveness of the proposed modules. Our source code is publicly available at https://github.com/WangXin81/OUNet_JL .

PMID:40025128 | DOI:10.1038/s41598-025-91839-7

Pest Manag Sci. 2025 Feb 28. doi: 10.1002/ps.8728. Online ahead of print.

ABSTRACT

BACKGROUND: Precision weed mapping in turf according to its susceptibility to selective herbicides allows the smart sprayer to spot-spray the most pertinent herbicides onto the susceptible weeds. The objective of this study was to evaluate the feasibility of implementing herbicide susceptibility-based weed mapping using deep convolutional neural networks (DCNNs) to facilitate targeted and efficient herbicide applications. Additionally, applying path-planning algorithms to weed mapping data to guide the spraying nozzle ensures minimal travel paths for herbicide application.

RESULTS: DenseNet achieved high precision, recall, overall accuracy, and F1 score values for all categories of herbicides and no herbicides, with F1 scores ranging from 0.996 to 0.999 in the validation dataset and from 0.992 to 0.997 in the testing dataset. The average accuracies attained by DenseNet, GoogLeNet and ResNet were 0.9985, 0.9953 and 0.9980, respectively. By considering both accuracy and computational efficiency, the ResNet model was identified as the most effective among the models compared to weed detection. The performance of the Christofides, Greedy and 2-opt algorithms in optimizing path planning for single or dual spraying nozzles was compared and analyzed. The Greedy algorithm proved the most efficient in optimizing the nozzle's trajectory.

CONCLUSION: Implementing herbicide susceptibility-based weed mapping facilitates targeted herbicide application by directing the nozzle to the grid cells containing the weeds susceptible to the herbicides. Moreover, the strategic integration of herbicide susceptibility-based weed mapping with optimized path planning for the spraying mechanism can be adeptly implemented on smart sprayers, which could effectively reduce the herbicide input. © 2025 Society of Chemical Industry.

PMID:40022516 | DOI:10.1002/ps.8728

J Cachexia Sarcopenia Muscle. 2025 Apr;16(2):e13728. doi: 10.1002/jcsm.13728.

ABSTRACT

BACKGROUND: Age-related conditions, such as osteoporosis and sarcopenia, alongside chronic diseases, can result in significant musculoskeletal tissue loss. This impacts individuals' quality of life and increases risk of falls and fractures. Computed tomography (CT) has been widely used for assessing musculoskeletal tissues. Although automatic techniques have been investigated for segmenting tissues in the abdomen and mid-thigh regions, studies in proximal hip remain limited. This study aims to develop a deep learning technique for segmentation and quantification of musculoskeletal tissues in CT scans of proximal hip.

METHODS: We examined 300 participants (men, 73 ± 6 years) from two cohorts of the Osteoporotic Fractures in Men Study (MrOS). We manually segmented cortical bone, trabecular bone, marrow adipose tissue (MAT), haematopoietic bone marrow (HBM), muscle, intermuscular adipose tissue (IMAT) and subcutaneous adipose tissue (SAT) from CT scan images at the proximal hip level. Using these data, we trained a U-Net-like deep learning model for automatic segmentation. The association between model-generated quantitative results and outcome variables such as grip strength, chair sit-to-stand time, walking speed, femoral neck and spine bone mineral density (BMD), and total lean mass was calculated.

RESULTS: An average Dice similarity coefficient (DSC) above 90% was observed across all tissue types in the test dataset. Grip strength showed positive correlations with cortical bone area (coefficient: 0.95, 95% confidence interval: [0.10, 1.80]), muscle area (0.41, [0.19, 0.64]) and average Hounsfield unit for muscle adjusted for height squared (AHU/h2) (1.1, [0.53, 1.67]), while it was negatively correlated with IMAT (-1.45, [-2.21, -0.70]) and SAT (-0.32, [-0.50, -0.13]). Gait speed was directly related to muscle area (0.01, [0.00, 0.02]) and inversely to IMAT (-0.04, [-0.07, -0.01]), while chair sit-to-stand time was associated with muscle area (0.98, [0.98, 0.99]), IMAT area (1.04, [1.01, 1.07]), SAT area (1.01, [1.01, 1.02]) and AHU/h2 for muscle (0.97, [0.95, 0.99]). MAT area showed a potential link to non-trauma fractures post-50 years (1.67, [0.98, 2.83]). Femoral neck BMD was associated with cortical bone (0.09, [0.08, 0.10]), MAT (-0.11, [-0.13, -0.10]), MAT adjusted for total bone marrow area (-0.06, [-0.07, -0.05]) and AHU/h2 for muscle (0.01, [0.00, 0.02]). Total spine BMD showed similar associations and with AHU for muscle (0.02, [0.00, 0.05]). Total lean mass was correlated with cortical bone (517.3, [148.26, 886.34]), trabecular bone (924, [262.55, 1585.45]), muscle (381.71, [291.47, 471.96]), IMAT (-1096.62, [-1410.34, -782.89]), SAT (-413.28, [-480.26, -346.29]), AHU (527.39, [159.12, 895.66]) and AHU/h2 (300.03, [49.23, 550.83]).

CONCLUSION: Our deep learning-based technique offers a fast and accurate method for segmentation and quantification of musculoskeletal tissues in proximal hip, with potential clinical value.

PMID:40022453 | DOI:10.1002/jcsm.13728

Food Res Int. 2025 Feb;203:115905. doi: 10.1016/j.foodres.2025.115905. Epub 2025 Feb 1.

ABSTRACT

Current detection methods for lipid and protein oxidation using hyperspectral imaging (HSI) in conjunction with machine learning (ML) necessitate the involvement of data scientists and domain experts to adjust the model architecture and tune hyperparameters. Additionally, prediction models lack explainability in the predictive outcomes and decision-making process. In this study, ML, automated machine learning (AutoML) and automated deep learning (AutoDL) models were developed for visible near-infrared HSI of mutton samples treated with different freeze-thaw cycles to evaluate the feasibility of building prediction models for lipid and protein oxidation without manual intervention. SHapley Additive exPlanations (SHAP) were utilized to explain the prediction models. The results showed that the AutoDL attained the effective prediction models for lipid oxidation (R2p = 0.9021, RMSEP = 0.0542 mg/kg, RPD = 3.3624) and protein oxidation (R2p = 0.8805, RMSEP = 3.8065 nmol/mg, RPD = 3.0789). AutoML driven stacked ensembles further improved the generalization ability of the models, predicting lipid and protein oxidation with R2p of 0.9237 and 0.9347. The important wavelengths identified through SHAP closely align with the results obtained from spectral analysis, and the analysis also determined the magnitude and direction of the impact of these important wavelengths on the model outputs. Finally, changes in lipid and protein oxidation of mutton in different freeze-thaw cycles were visualized. The research indicated that the combination of HSI, AutoML and SHAP may generate high-quality explainable models without human assistance for monitoring lipid and protein oxidative damage in mutton.

PMID:40022412 | DOI:10.1016/j.foodres.2025.115905

Food Res Int. 2025 Feb;203:115874. doi: 10.1016/j.foodres.2025.115874. Epub 2025 Jan 30.

ABSTRACT

Fixation is a critical step in green tea processing, and the moisture content of the leaves after fixation is a key indicator of the fixation quality. Near-infrared spectroscopy (NIRS)-based moisture detection technology is often applied in the tea processing industry. However, temperature fluctuations during processing can cause changes in the NIRS curves, which in turn affect the accuracy of moisture prediction models based on the spectral data. To address this challenge, NIRS data were collected from samples at various stages of fixation and at different temperatures, and a novel deep learning network (DiSENet) was proposed, which integrates multi-scale feature fusion and attention mechanisms. Using a global modeling approach, the proposed method achieved a coefficient of determination (RP2) of 0.781 for moisture content prediction, with a root mean square error (RMSEP) of 1.720 % and a residual predictive deviation (RPD) of 2.148. On the dataset constructed for this study, DiSENet demonstrated superior predictive accuracy compared to the spectral correction methods of external parameter orthogonalization (EPO) and generalized least squares weighting (GLSW), as well as traditional global modeling methods such as partial least squares regression (PLSR) and support vector regression (SVR). This approach effectively corrects spectral interferences caused by temperature variations, thereby enhancing the accuracy of moisture content prediction. Thus, it offers a reliable solution for real-time, non-destructive moisture detection during tea processing.

PMID:40022390 | DOI:10.1016/j.foodres.2025.115874

Food Res Int. 2025 Feb;203:115810. doi: 10.1016/j.foodres.2025.115810. Epub 2025 Jan 22.

ABSTRACT

The application of high-level data fusion in the detection of agricultural products still presents a significant challenge. In this study, dual-channel feature fusion model (DCFFM) with attention mechanism was proposed to optimize the utilization of both one-dimensional spectral data and two-dimensional image data in the hyperspectral images for achieving high-level data fusion. A comparative analysis of support vector machine (SVM), convolutional neural network (CNN) with DCFFM, demonstrated that DCFFM exhibited superior results, achieving the accuracy, precision, recall, specificity, and F1-score of 95.13 %, 95.49 %, 94.83 %, 98.97 %, 95.12 % in the visible and near-infrared (Vis-NIR), and 94.00 %, 94.43 %, 94.16 %, 98.67 %, 94.27 % in the short-wave infrared (SWIR). This also indicated that Vis-NIR was more suitable for identifying unsound soybeans than SWIR. Furthermore, visualization was employed to demonstrate classification outcomes, thereby illustrating the generalization capacity of DCFFM through model inversion. In summary, this study is to explore a modeling framework that is capable of the comprehensive acquisition of spectra and images in the hyperspectral images, allowing for high-level data fusion, thereby achieving enhanced levels of accuracy.

PMID:40022337 | DOI:10.1016/j.foodres.2025.115810

Cancer Imaging. 2025 Feb 28;25(1):20. doi: 10.1186/s40644-025-00845-5.

ABSTRACT

OBJECTIVE: Exploring the construction of a fusion model that combines radiomics and deep learning (DL) features is of great significance for the precise preoperative diagnosis of meningioma sinus invasion.

MATERIALS AND METHODS: This study retrospectively collected data from 601 patients with meningioma confirmed by surgical pathology. For each patient, 3948 radiomics features, 12,288 VGG features, 6144 ResNet features, and 3072 DenseNet features were extracted from MRI images. Thus, univariate logistic regression, correlation analysis, and the Boruta algorithm were applied for further feature dimension reduction, selecting radiomics and DL features highly associated with meningioma sinus invasion. Finally, diagnosis models were constructed using the random forest (RF) algorithm. Additionally, the diagnostic performance of different models was evaluated using receiver operating characteristic (ROC) curves, and AUC values of different models were compared using the DeLong test.

RESULTS: Ultimately, 21 features highly associated with meningioma sinus invasion were selected, including 6 radiomics features, 2 VGG features, 7 ResNet features, and 6 DenseNet features. Based on these features, five models were constructed: the radiomics model, VGG model, ResNet model, DenseNet model, and DL-radiomics (DLR) fusion model. This fusion model demonstrated superior diagnostic performance, with AUC values of 0.818, 0.814, and 0.769 in the training set, internal validation set, and independent external validation set, respectively. Furthermore, the results of the DeLong test indicated that there were significant differences between the fusion model and both the radiomics model and the VGG model (p < 0.05).

CONCLUSIONS: The fusion model combining radiomics and DL features exhibits superior diagnostic performance in preoperative diagnosis of meningioma sinus invasion. It is expected to become a powerful tool for clinical surgical plan selection and patient prognosis assessment.

PMID:40022261 | DOI:10.1186/s40644-025-00845-5

Genome Med. 2025 Feb 28;17(1):16. doi: 10.1186/s13073-025-01442-8.

ABSTRACT

Spatial transcriptomics (ST) enables the study of gene expression in spatial context, but many ST technologies face challenges due to limited resolution, leading to cell mixtures at each spot. We present LETSmix to deconvolve cell types by integrating spatial correlations through a tailored LETS filter, which leverages layer annotations, expression similarities, image texture features, and spatial coordinates to refine ST data. Additionally, LETSmix employs a mixup-augmented domain adaptation strategy to address discrepancies between ST and reference single-cell RNA sequencing data. Comprehensive evaluations across diverse ST platforms and tissue types demonstrate its high accuracy in estimating cell-type proportions and spatial patterns, surpassing existing methods (URL: https://github.com/ZhanYangen/LETSmix ).

PMID:40022231 | DOI:10.1186/s13073-025-01442-8