Sci Rep. 2025 Mar 22;15(1):9942. doi: 10.1038/s41598-024-82909-3.

ABSTRACT

To develop a deep learning (DL) model based on MRI to predict muscle-invasive bladder cancer (MIBC). A total of 559 patients, including 521 patients in our center and 38 patients in external centers were collected from 2012 to 2023 to construct the DL model. In this study, the DL model was utilized to differentiate between MIBC and NMIBC based on three-channel image inputs, including original T2WI images, segmented bladder, and regions of interest. Inception V3 was employed for model construction. The accuracy, sensitivity (SN), specificity (SP), positive predictive value (PPV) and negative predictive value (NPV) for predicting MIBC by DL model were 92.4%, 94.7%, 91.5%, 81.8% and 97.7% in the validation set and 92.1%, 86.8%, 94.6%, 88.5% and 93.8% in the internal test set. In the external test set, these values were 81.6%, 57.1%, 87.1%, 50.0% and 90.0%. Additionally, the accuracy, SN, SP, PPV, and NPV for predicting MIBC were 93.5%, 100%, 93.4%, 11.1%, and 100% in VI-RADS 2; 80.0%, 66.7%, 87.2%, 73.7% and 82.9% in VI-RADS 3; 90.3%, 91.7%, 85.7%, 95.7%, 75.0% in VI-RADS 4. The accuracy, SN, and PPV were 93.9%, 93.9%, and 100% in VI-RADS 5. The DL model based on T2WI can effectively predict MIBC and serve as a valuable complement to VI-RADS 3.

PMID:40121216 | DOI:10.1038/s41598-024-82909-3

BJS Open. 2025 Mar 4;9(2):zraf013. doi: 10.1093/bjsopen/zraf013.

ABSTRACT

BACKGROUND: A real-time deep learning system was developed to identify the extrahepatic bile ducts during indocyanine green fluorescence-guided laparoscopic cholecystectomy.

METHODS: Two expert surgeons annotated surgical videos from 113 patients and six class structures. YOLOv7, a real-time object detection model that enhances speed and accuracy in identifying and localizing objects within images, was trained for structures identification. To evaluate the model's performance, single-frame and short video clip validations were used. The primary outcomes were average precision and mean average precision in single-frame validation. Secondary outcomes were accuracy and other metrics in short video clip validations. An intraoperative prototype was developed for the verification experiments.

RESULTS: A total of 3993 images were extracted to train the YOLOv7 model. In single-frame validation, all classes' mean average precision was 0.846, and average precision for the common bile duct and cystic duct was 0.864 and 0.698 respectively. The model was trained to detect six different classes of objects and exhibited the best overall performance, with an accuracy of 94.39% for the common bile duct and 84.97% for the cystic duct in video clip validation.

CONCLUSION: This model could potentially assist surgeons in identifying the critical landmarks during laparoscopic cholecystectomy, thereby minimizing the risk of bile duct injuries.

PMID:40119711 | DOI:10.1093/bjsopen/zraf013

IET Syst Biol. 2025 Jan-Dec;19(1):e70010. doi: 10.1049/syb2.70010.

ABSTRACT

Cancer is a serious and complex disease caused by uncontrolled cell growth and is becoming one of the leading causes of death worldwide. Anticancer peptides (ACPs), as a bioactive peptide with lower toxicity, emerge as a promising means of effectively treating cancer. Identifying ACPs is challenging due to the limitation of experimental conditions. To address this, we proposed a dual-channel-based deep learning method, termed ACP-DPE, for ACP prediction. The ACP-DPE consisted of two parallel channels: one was an embedding layer followed by the bi-directional gated recurrent unit (Bi-GRU) module, and the other was an adaptive embedding layer followed by the dilated convolution module. The Bi-GRU module captured the peptide sequence dependencies, whereas the dilated convolution module characterised the local relationship of amino acids. Experimental results show that ACP-DPE achieves an accuracy of 82.81% and a sensitivity of 86.63%, surpassing the state-of-the-art method by 3.86% and 5.1%, respectively. These findings demonstrate the effectiveness of ACP-DPE for ACP prediction and highlight its potential as a valuable tool in cancer treatment research.

PMID:40119615 | DOI:10.1049/syb2.70010

Pest Manag Sci. 2025 Mar 22. doi: 10.1002/ps.8790. Online ahead of print.

ABSTRACT

BACKGROUND: Rice diseases that are not detected in a timely manner may trigger large-scale yield reduction and bring significant economic losses to farmers.

AIMS: In order to solve the problems of insufficient rice disease detection accuracy and a model that is lightweight, this study proposes a lightweight rice disease detection method based on the improved YOLOv8. The method incorporates a full-dimensional dynamic convolution (ODConv) module to enhance the feature extraction capability and improve the robustness of the model, while a dynamic non-monotonic focusing mechanism, WIoU (weighted interpolation of sequential evidence for intersection over union), is employed to optimize the bounding box loss function for faster convergence and improved detection performance. In addition, the use of a high-resolution detector head improves the small target detection capability and reduces the network parameters by removing redundant layers.

RESULTS: Experimental results show a 66.6% reduction in parameters and a 61.9% reduction in model size compared to the YOLOv8n baseline. The model outperforms Faster R-CNN, YOLOv5s, YOLOv6n, YOLOv7-tiny, and YOLOv8n by 29.2%, 3.8%, 5.2%, 5.7%, and 5.2%, respectively, in terms of the mean average precision (mAP), which shows a significant improvement in the detection performance.

CONCLUSION: The YOLOv8-OW model provides a more effective solution, which is suitable for deployment on resource-limited mobile devices, to provide real-time and accurate disease detection support for farmers and further promotes the development of precision agriculture. © 2025 Society of Chemical Industry.

PMID:40119571 | DOI:10.1002/ps.8790

Genome Biol. 2025 Mar 21;26(1):64. doi: 10.1186/s13059-025-03519-4.

ABSTRACT

Discovering a lower-dimensional embedding of single-cell data can improve downstream analysis. The embedding should encapsulate both the high-level features and low-level variations. While existing generative models attempt to learn such low-dimensional representations, they have limitations. Here, we introduce scVAEDer, a scalable deep-learning model that combines the power of variational autoencoders and deep diffusion models to learn a meaningful representation that retains both global structure and local variations. Using the learned embeddings, scVAEDer can generate novel scRNA-seq data, predict perturbation response on various cell types, identify changes in gene expression during dedifferentiation, and detect master regulators in biological processes.

PMID:40119479 | DOI:10.1186/s13059-025-03519-4

J Cheminform. 2025 Mar 21;17(1):35. doi: 10.1186/s13321-025-00979-5.

ABSTRACT

Protein-protein interactions (PPIs) are central to the mechanisms of signaling pathways and immune responses, which can help us understand disease etiology. Therefore, there is a significant need for efficient and rapid automated approaches to predict changes in PPIs. In recent years, there has been a significant increase in applying deep learning techniques to predict changes in binding affinity between the original protein complex and its mutant variants. Particularly, the adoption of graph neural networks (GNNs) has gained prominence for their ability to learn representations of protein-protein complexes. However, the conventional GNNs have mainly concentrated on capturing local features, often disregarding the interactions among distant elements that hold potential important information. In this study, we have developed a transformer-based graph neural network to extract features of the mutant segment from the three-dimensional structure of protein-protein complexes. By embracing both local and global features, the approach ensures a more comprehensive understanding of the intricate relationships, thus promising more accurate predictions of binding affinity changes. To enhance the representation capability of protein features, we incorporate a large-scale pre-trained protein language model into our approach and employ the global protein feature it provides. The proposed model is shown to be able to predict the mutation changes in binding affinity with a root mean square error of 1.10 and a Pearson correlation coefficient of near 0.71, as demonstrated by performance on test and validation cases. Our experiments on all five datasets, including both single mutant and multiple mutant cases, demonstrate that our model outperforms four state-of-the-art baseline methods, and the efficacy was subjected to comprehensive experimental evaluation. Our study introduces a transformer-based graph neural network approach to accurately predict changes in protein-protein interactions (PPIs). By integrating local and global features and leveraging pretrained protein language models, our model outperforms state-of-the-art methods across diverse datasets. The results of this study can provide new views for studying immune responses and disease etiology related to protein mutations. Furthermore, this approach may contribute to other biological or biochemical studies related to PPIs.Scientific contribution Our scientific contribution lies in the development of a novel transformer-based graph neural network tailored to predict changes in protein-protein interactions (PPIs) with excellent accuracy. By seamlessly integrating both local and global features extracted from the three-dimensional structure of protein-protein complexes, and leveraging the rich representations provided by pretrained protein language models, our approach surpasses existing methods across diverse datasets. Our findings may offer novel insights for the understanding of complex disease etiology associated with protein mutations. The novel tool can be applicable to various biological and biochemical investigations involving protein mutations.

PMID:40119464 | DOI:10.1186/s13321-025-00979-5

Cancer Cell Int. 2025 Mar 21;25(1):111. doi: 10.1186/s12935-025-03756-4.

ABSTRACT

Nowadays, gastric cancer has become a significant issue in the global cancer burden, and its impact cannot be ignored. The rapid development of artificial intelligence technology is attempting to address this situation, aiming to change the clinical management landscape of gastric cancer fundamentally. In this transformative change, machine learning and deep learning, as two core technologies, play a pivotal role, bringing unprecedented innovations and breakthroughs in the diagnosis, treatment, and prognosis evaluation of gastric cancer. This article comprehensively reviews the latest research status and application of artificial intelligence algorithms in gastric cancer, covering multiple dimensions such as image recognition, pathological analysis, personalized treatment, and prognosis risk assessment. These applications not only significantly improve the sensitivity of gastric cancer risk monitoring, the accuracy of diagnosis, and the precision of survival prognosis but also provide robust data support and a scientific basis for clinical decision-making. The integration of artificial intelligence, from optimizing the diagnosis process and enhancing diagnostic efficiency to promoting the practice of precision medicine, demonstrates its promising prospects for reshaping the treatment model of gastric cancer. Although most of the current AI-based models have not been widely used in clinical practice, with the continuous deepening and expansion of precision medicine, we have reason to believe that a new era of AI-driven gastric cancer care is approaching.

PMID:40119433 | DOI:10.1186/s12935-025-03756-4

BMC Med Imaging. 2025 Mar 21;25(1):95. doi: 10.1186/s12880-025-01636-x.

ABSTRACT

PURPOSE: To develop an automatic segmentation model for surgical marks, titanium clips, in target volume delineation of breast cancer radiotherapy after lumpectomy.

METHODS: A two-stage deep-learning model is used to segment the titanium clips from CT image. The first network, Location Net, is designed to search the region containing all clips from CT. Then the second network, Segmentation Net, is designed to search the locations of clips from the previously detected region. Ablation studies are performed to evaluate the impact of various inputs for both networks. The two-stage deep-learning model is also compared with the other existing deep-learning methods including U-Net, V-Net and UNETR. The segmentation accuracy of these models is evaluated by three metrics: Dice Similarity Coefficient (DSC), 95% Hausdorff Distance (HD95), and Average Surface Distance (ASD).

RESULTS: The DSC, HD95 and ASD of the two-stage model are 0.844, 2.008 mm and 0.333 mm, while their values are 0.681, 2.494 mm and 0.785 mm for U-Net, 0.767, 2.331 mm and 0.497 mm for V-Net, 0.714, 2.660 mm and 0.772 mm for UNETR. The proposed 2-stage model achieved the best performance among the four models.

CONCLUSION: With the two-stage searching strategy the accuracy to detect titanium clips can be improved comparing to those existing deep-learning models with one-stage searching strategy. The proposed segmentation model can facilitate the delineation of tumor bed and subsequent target volume for breast cancer radiotherapy after lumpectomy.

PMID:40119258 | DOI:10.1186/s12880-025-01636-x

Intensive Care Med Exp. 2025 Mar 21;13(1):39. doi: 10.1186/s40635-025-00746-8.

ABSTRACT

BACKGROUND: Patient-ventilator asynchrony (PVA) is a mismatch between the patient's respiratory drive/effort and the ventilator breath delivery. It occurs frequently in mechanically ventilated patients and has been associated with adverse events and increased duration of ventilation. Identifying PVA through visual inspection of ventilator waveforms is highly challenging and time-consuming. Automated PVA detection using Artificial Intelligence (AI) has been increasingly studied, potentially offering real-time monitoring at the bedside. In this review, we discuss advances in automatic detection of PVA, focusing on developments of the last 15 years.

RESULTS: Nineteen studies were identified. Multiple forms of AI have been used for the automated detection of PVA, including rule-based algorithms, machine learning and deep learning. Three licensed algorithms are currently reported. Results of algorithms are generally promising (average reported sensitivity, specificity and accuracy of 0.80, 0.93 and 0.92, respectively), but most algorithms are only available offline, can detect a small subset of PVAs (focusing mostly on ineffective effort and double trigger asynchronies), or remain in the development or validation stage (84% (16/19 of the reviewed studies)). Moreover, only in 58% (11/19) of the studies a reference method for monitoring patient's breathing effort was available. To move from bench to bedside implementation, data quality should be improved and algorithms that can detect multiple PVAs should be externally validated, incorporating measures for breathing effort as ground truth. Last, prospective integration and model testing/finetuning in different ICU settings is key.

CONCLUSIONS: AI-based techniques for automated PVA detection are increasingly studied and show potential. For widespread implementation to succeed, several steps, including external validation and (near) real-time employment, should be considered. Then, automated PVA detection could aid in monitoring and mitigating PVAs, to eventually optimize personalized mechanical ventilation, improve clinical outcomes and reduce clinician's workload.

PMID:40119215 | DOI:10.1186/s40635-025-00746-8

Sci Rep. 2025 Mar 21;15(1):9845. doi: 10.1038/s41598-025-94664-0.

ABSTRACT

Colorectal cancer (CRC) ranks as the third most common and second deadliest cancer worldwide. The immune system, particularly tertiary lymphoid structures (TLS), significantly influences CRC progression and prognosis. TLS maturation, especially in the presence of germinal centers, correlates with improved patient outcomes; however, consistent and objective TLS assessment is hindered by varying histological definitions and limitations of traditional staining methods. This study involved 656 patients with colorectal adenocarcinoma from CHU Brest, France. We employed dual immunohistochemistry staining for CD21 and CD23 to classify TLS maturation stages in whole-slide images and implemented a fivefold cross-validation. Using ResNet50 and Vision Transformer models, we compared various aggregation methods, architectures, and pretraining techniques. Our automated system, TLS-PAT, achieved high accuracy (0.845) and robustness (kappa = 0.761) in classifying TLS maturation, particularly with the Vision Transformer pretrained on ImageNet using Max Confidence aggregation. This AI-driven approach offers a standardized method for automated TLS classification, complementing existing detection techniques. Our open-source tools are designed for easy integration with current methods, paving the way for further research in external datasets and other cancer types.

PMID:40119179 | DOI:10.1038/s41598-025-94664-0

Sci Rep. 2025 Mar 21;15(1):9805. doi: 10.1038/s41598-025-94680-0.

ABSTRACT

Accurate embryo morphology assessment is essential in assisted reproductive technology for selecting the most viable embryo. Artificial intelligence has the potential to enhance this process. However, the limited availability of embryo data presents challenges for training deep learning models. To address this, we trained two generative models using two datasets-one we created and made publicly available, and one existing public dataset-to generate synthetic embryo images at various cell stages, including 2-cell, 4-cell, 8-cell, morula, and blastocyst. These were combined with real images to train classification models for embryo cell stage prediction. Our results demonstrate that incorporating synthetic images alongside real data improved classification performance, with the model achieving 97% accuracy compared to 94.5% when trained solely on real data. This trend remained consistent when tested on an external Blastocyst dataset from a different clinic. Notably, even when trained exclusively on synthetic data and tested on real data, the model achieved a high accuracy of 92%. Furthermore, combining synthetic data from both generative models yielded better classification results than using data from a single generative model. Four embryologists evaluated the fidelity of the synthetic images through a Turing test, during which they annotated inaccuracies and offered feedback. The analysis showed the diffusion model outperformed the generative adversarial network, deceiving embryologists 66.6% versus 25.3% and achieving lower Fréchet inception distance scores.

PMID:40119109 | DOI:10.1038/s41598-025-94680-0

Radiother Oncol. 2025 Mar 19:110852. doi: 10.1016/j.radonc.2025.110852. Online ahead of print.

ABSTRACT

BACKGROUND: In the HECKTOR 2022 challenge set [1], several state-of-the-art (SOTA, achieving best performance) deep learning models were introduced for predicting recurrence-free period (RFP) in head and neck cancer patients using PET and CT images.

PURPOSE: This study investigates whether a conventional DenseNet architecture, with optimized numbers of layers and image-fusion strategies, could achieve comparable performance as SOTA models.

METHODS: The HECKTOR 2022 dataset comprises 489 oropharyngeal cancer (OPC) patients from seven distinct centers. It was randomly divided into a training set (n = 369) and an independent test set (n = 120). Furthermore, an additional dataset of 400 OPC patients, who underwent chemo(radiotherapy) at our center, was employed for external testing. Each patients' data included pre-treatment CT- and PET-scans, manually generated GTV (Gross tumour volume) contours for primary tumors and lymph nodes, and RFP information. The present study compared the performance of DenseNet against three SOTA models developed on the HECKTOR 2022 dataset.

RESULTS: When inputting CT, PET and GTV using the early fusion (considering them as different channels of input) approach, DenseNet81 (with 81 layers) obtained an internal test C-index of 0.69, a performance metric comparable with SOTA models. Notably, the removal of GTV from the input data yielded the same internal test C-index of 0.69 while improving the external test C-index from 0.59 to 0.63. Furthermore, compared to PET-only models, when utilizing the late fusion (concatenation of extracted features) with CT and PET, DenseNet81 demonstrated superior C-index values of 0.68 and 0.66 in both internal and external test sets, while using early fusion was better in only the internal test set.

CONCLUSIONS: The basic DenseNet architecture with 81 layers demonstrated a predictive performance on par with SOTA models featuring more intricate architectures in the internal test set, and better performance in the external test. The late fusion of CT and PET imaging data yielded superior performance in the external test.

PMID:40118186 | DOI:10.1016/j.radonc.2025.110852

Food Chem Toxicol. 2025 Mar 19:115401. doi: 10.1016/j.fct.2025.115401. Online ahead of print.

ABSTRACT

Acrylamide (AA), a food hazard generated during thermal processing, poses significant safety risks due to its toxicity. Conventional methods for AA toxicology are time-consuming and inadequate for analyzing cellular morphology. This study developed a novel approach combining deep learning models (U-Net and ResNet34) with cell fluorescence imaging. U-Net was used for cell segmentation, generating a single-cell dataset, while ResNet34 trained the dataset over 200 epochs, achieving an 80% validation accuracy. This method predicts AA concentration ranges by matching cell fluorescence features with the dataset and analyzes cellular phenotypic changes under AA exposure using k-means clustering and CellProfiler. The approach overcomes the limitations of traditional toxicological methods, offering a direct link between cell phenotypes and hazard toxicology. It provides a high-throughput, accurate solution to evaluate AA toxicology and refines the understanding of its cellular impacts.

PMID:40118138 | DOI:10.1016/j.fct.2025.115401

Cell Rep Med. 2025 Mar 14:102032. doi: 10.1016/j.xcrm.2025.102032. Online ahead of print.

ABSTRACT

Liver tumors, whether primary or metastatic, significantly impact the outcomes of patients with cancer. Accurate identification and quantification are crucial for effective patient management, including precise diagnosis, prognosis, and therapy evaluation. We present SALSA (system for automatic liver tumor segmentation and detection), a fully automated tool for liver tumor detection and delineation. Developed on 1,598 computed tomography (CT) scans and 4,908 liver tumors, SALSA demonstrates superior accuracy in tumor identification and volume quantification, outperforming state-of-the-art models and inter-reader agreement among expert radiologists. SALSA achieves a patient-wise detection precision of 99.65%, and 81.72% at lesion level, in the external validation cohorts. Additionally, it exhibits good overlap, achieving a dice similarity coefficient (DSC) of 0.760, outperforming both state-of-the-art and the inter-radiologist assessment. SALSA's automatic quantification of tumor volume proves to have prognostic value across various solid tumors (p = 0.028). SALSA's robust capabilities position it as a potential medical device for automatic cancer detection, staging, and response evaluation.

PMID:40118052 | DOI:10.1016/j.xcrm.2025.102032

Neural Netw. 2025 Mar 15;187:107414. doi: 10.1016/j.neunet.2025.107414. Online ahead of print.

ABSTRACT

The identification of early micro-lesions and adjacent blood vessels in CT scans plays a pivotal role in the clinical diagnosis of pancreatic cancer, considering its aggressive nature and high fatality rate. Despite the widespread application of deep learning methods for this task, several challenges persist: (1) the complex background environment in abdominal CT scans complicates the accurate localization of potential micro-tumors; (2) the subtle contrast between micro-lesions within pancreatic tissue and the surrounding tissues makes it challenging for models to capture these features accurately; and (3) tumors that invade adjacent blood vessels pose significant barriers to surgical procedures. To address these challenges, we propose LUNETR (Language-Infused UNETR), an advanced multimodal encoder model that combines textual and image information for precise medical image segmentation. The integration of an autoencoding language model with cross-attention enabling our model to effectively leverage semantic associations between textual and image data, thereby facilitating precise localization of potential pancreatic micro-tumors. Additionally, we designed a Multi-scale Aggregation Attention (MSAA) module to comprehensively capture both spatial and channel characteristics of global multi-scale image data, enhancing the model's capacity to extract features from micro-lesions embedded within pancreatic tissue. Furthermore, in order to facilitate precise segmentation of pancreatic tumors and nearby blood vessels and address the scarcity of multimodal medical datasets, we collaborated with Zhuzhou Central Hospital to construct a multimodal dataset comprising CT images and corresponding pathology reports from 135 pancreatic cancer patients. Our experimental results surpass current state-of-the-art models, with the incorporation of the semantic encoder improving the average Dice score for pancreatic tumor segmentation by 2.23 %. For the Medical Segmentation Decathlon (MSD) liver and lung cancer datasets, our model achieved an average Dice score improvement of 4.31 % and 3.67 %, respectively, demonstrating the efficacy of the LUNETR.

PMID:40117980 | DOI:10.1016/j.neunet.2025.107414

Mar Pollut Bull. 2025 Mar 19;215:117813. doi: 10.1016/j.marpolbul.2025.117813. Online ahead of print.

ABSTRACT

Efficiently monitoring red tide via satellite remote sensing is pivotal for marine disaster monitoring and ecological early warning systems. Traditional remote sensing methods for monitoring red tide typically rely on ocean colour sensors with low spatial resolution and high spectral resolution, making it difficult to monitor small events and detailed distribution of red tide. Furthermore, traditional methods are not applicable to satellite sensors with medium to high spatial resolution and low spectral resolution, significantly limiting the ability to detect red tide outbreaks in their early stages. Therefore, this study proposes a Residual Neural Network Red Tide Monitoring Model based on Spectral Information Channel Constraints (SIC-RTNet) using HY-1C/D CZI satellite data. SIC-RTNet improves monitoring accuracy through adding three key steps compared to basic deep learning methods. First, the SIC-RTNet introduces residual blocks to enhance the effective retention and transmission of weak surface signal features of red tides. Second, three spectral information channels are calculated using the four wideband channels of the images to amplify the spectral differences between red tide and seawater. Finally, an improved loss function is employed to address the issue of sample imbalance between red tides and seawater. Compared to other models, SIC-RTNet demonstrates superior performance, achieving precision and recall rates of 85.5 % and 95.4 % respectively. The F1-Score is 0.90, and the Mean Intersection over Union (MoU) is 0.90. The results indicate that the SIC-RTNet can automatically identify red tides using high spatial resolution and wideband remote sensing data, which can help the monitoring of marine ecological disasters.

PMID:40117936 | DOI:10.1016/j.marpolbul.2025.117813

Food Chem. 2025 Mar 17;480:143854. doi: 10.1016/j.foodchem.2025.143854. Online ahead of print.

ABSTRACT

Cereals are a primary source of sustenance for humanity. Monitoring, controlling, and preventing mycotoxins in cereals are vital for ensuring the safety of the cereals and their derived products. This study introduces transfer learning strategies into chemometrics to improve deep learning models applied to spectral data from different grains or toxins. Three transfer learning methods were explored for their potential to quantitatively detect fungal toxins in cereals. The feasibility of transfer learning was demonstrated by predicting wheat zearalenone (ZEN) and peanut aflatoxin B1 (AFB1) sample sets on different instruments. The results indicated that the second transfer method is effective in detecting toxins. For FT-NIR spectrometry, the transfer model achieved an R2 of 0.9356, a relative prediction deviation (RPD) of 3.9497 for wheat ZEN prediction, and an R2 of 0.9419 with an RPD of 4.1551 for peanut AFB1 detection. With NIR spectrometry, effective peanut AFB1 detection was also achieved, yielding an R2 of 0.9386 and an RPD of 4.0434 in the prediction set. These results suggest that the proposed transfer learning approach can successfully update a source domain model into one that is suitable for tasks in the target domain. This study provides a viable solution to the problem of poor adaptability of single-source models, presenting a more universally applicable method for spectral detection of fungal toxins in cereals.

PMID:40117813 | DOI:10.1016/j.foodchem.2025.143854

Comput Biol Med. 2025 Mar 20;190:110007. doi: 10.1016/j.compbiomed.2025.110007. Online ahead of print.

ABSTRACT

Skin diseases affect over a third of the global population, yet their impact is often underestimated. Automating the classification of these diseases is essential for supporting timely and accurate diagnoses. This study leverages Vision Transformers, Swin Transformers, and DinoV2, introducing DinoV2 for the first time in dermatology tasks. On a 31-class skin disease dataset, DinoV2 achieves state-of-the-art results with a test accuracy of 96.48 ± 0.0138% and an F1-Score of 97.27%, marking a nearly 10% improvement over existing benchmarks. The robustness of DinoV2 is further validated on the HAM10000 and Dermnet datasets, where it consistently surpasses prior models. Comparative analysis also includes ConvNeXt and other CNN architectures, underscoring the benefits of transformer models. Additionally, explainable AI techniques like GradCAM and SHAP provide global heatmaps and pixel-level correlation plots, offering detailed insights into disease localization. These complementary approaches enhance model transparency and support clinical correlations, assisting dermatologists in accurate diagnosis and treatment planning. This combination of high performance and clinical relevance highlights the potential of transformers, particularly DinoV2, in dermatological applications.

PMID:40117795 | DOI:10.1016/j.compbiomed.2025.110007

Ultrasonics. 2025 Mar 14;152:107641. doi: 10.1016/j.ultras.2025.107641. Online ahead of print.

ABSTRACT

This study aims to explore the feasibility of a deep learning approach to correct the distortion caused by the skull, thereby developing a transcranial adaptive focusing method for safe ultrasonic treatment in opening of the blood-brain barrier (BBB). However, aberration correction often requires significant computing power and time to ensure the accuracy of phase correction. This is due to the need to solve the evolution procedure of the sound field represented by numerous discretized grids. A combined method is proposed to train the phase prediction model for correcting the phase accurately and quickly. The method comprises pre-segmentation, k-Wave simulation, and a 3D U-net-based network. We use the k-Wave toolbox to construct a nonlinear simulation environment consisting of a 256-element phased array, a small piece of skull, and water. The skull sound speed sample combining with the phase delay serves as input for the model training. The focus volume and grating lobe level obtained by the proposed approach were the closest to those obtained by the time reversal method in all relevant approaches. Furthermore, the mean peak value obtained by the proposed approach was no less than 77% of that of the time reversal method. In this study, the computational cost of each sample's phase delay was no more than 0.05 s, which was 1/200th of the time reversal method. The proposed method eliminates the complexity of numerical calculation processes requiring consideration of more acoustic parameters, while circumventing the substantial computational resource demands and time-consuming challenges to traditional numerical approaches. The proposed method enables rapid, precise, and adaptive transcranial aberration correction on the 3D skull-based conditions, overcoming the potential inaccuracies in predicting the focal position or the acoustic energy distribution from 2D simulations. These results show the possibility of the proposed approach enabling near-real-time correction of skull-induced phase aberrations to achieve transcranial focus, thereby offering a novel option for treating brain diseases through temporary BBB opening.

PMID:40117699 | DOI:10.1016/j.ultras.2025.107641

J Med Internet Res. 2025 Mar 21;27:e60148. doi: 10.2196/60148.

ABSTRACT

BACKGROUND: Despite the rapid growth of research in artificial intelligence/machine learning (AI/ML), little is known about how often study results are disclosed years after study completion.

OBJECTIVE: We aimed to estimate the proportion of AI/ML research that reported results through ClinicalTrials.gov or peer-reviewed publications indexed in PubMed or Scopus.

METHODS: Using data from the Clinical Trials Transformation Initiative Aggregate Analysis of ClinicalTrials.gov, we identified studies initiated and completed between January 2010 and December 2023 that contained AI/ML-specific terms in the official title, brief summary, interventions, conditions, detailed descriptions, primary outcomes, or keywords. For 842 completed studies, we searched PubMed and Scopus for publications containing study identifiers and AI/ML-specific terms in relevant fields, such as the title, abstract, and keywords. We calculated disclosure rates within 3 years of study completion and median times to disclosure-from the "primary completion date" to the "results first posted date" on ClinicalTrials.gov or the earliest date of journal publication.

RESULTS: Of 842 completed studies (n=357 interventional; n=485 observational), 5.5% (46/842) disclosed results on ClinicalTrials.gov, 13.9% (117/842) in journal publications, and 17.7% (149/842) through either route within 3 years of completion. Higher disclosure rates were observed for trials: 10.4% (37/357) on ClinicalTrials.gov, 19.3% (69/357) in journal publications, and 26.1% (93/357) through either route. Randomized controlled trials had even higher disclosure rates: 11.3% (23/203) on ClinicalTrials.gov, 24.6% (50/203) in journal publications, and 32% (65/203) through either route. Nevertheless, most study findings (82.3%; 693/842) remained undisclosed 3 years after study completion. Trials using randomization (vs nonrandomized) or masking (vs open label) had higher disclosure rates and shorter times to disclosure. Most trials (85%; 305/357) had sample sizes of ≤1000, yet larger trials (n>1000) had higher publication rates (30.8%; 16/52) than smaller trials (n≤1000) (17.4%; 53/305). Hospitals (12.4%; 42/340), academia (15.1%; 39/259), and industry (13.7%; 20/146) published the most. High-income countries accounted for 82.4% (89/108) of all published studies. Of studies with disclosed results, the median times to report through ClinicalTrials.gov and in journal publications were 505 days (IQR 399-676) and 407 days (IQR 257-674), respectively. Open-label trials were common (60%; 214/357). Single-center designs were prevalent in both trials (83.3%; 290/348) and observational studies (82.3%; 377/458).

CONCLUSIONS: For over 80% of AI/ML studies completed during 2010-2023, study findings remained undisclosed even 3 years after study completion, raising questions about the representativeness of publicly available evidence. While methodological rigor was generally associated with higher publication rates, the predominance of single-center designs and high-income countries may limit the generalizability of the results currently accessible.

PMID:40117574 | DOI:10.2196/60148