Network. 2025 Apr 17:1-37. doi: 10.1080/0954898X.2025.2486206. Online ahead of print.

ABSTRACT

Brain Tumor (BT) is an irregular growth of cells in the brain or in the tissues surrounding it. Detecting and predicting tumours is essential in today's world, yet managing these diseases poses a considerable challenge. Among the various modalities, Magnetic Resonance Imaging (MRI) has been extensively exploited for diagnosing tumours. The traditional methods for predicting survival are based on handcrafted features from MRI and clinical information, which is generally subjective and laborious. This paper devises a new method named, Deep Residual PyramidNet (DRP_Net) for BT classification and survival prediction. The input MRI image is primarily derived from the BraTS dataset. Then, image enhancement is done to improve the quality of images using homomorphic filtering. Next, deep joint segmentation is used to process the tumourtumour region segmentation. Consequently, Haar wavelet and Local Directional Number Pattern (LDNP) based feature extraction is mined. Afterward, BT classification is achieved through DRP_Net, which is a fusion of Deep Residual Network (DRN) and PyramidNet. At last, the survival prediction is accomplished by employing the Deep Recurrent Neural Network (DRNN). Furthermore, DRP_Net has attained superior performance with a True Negative Rate (TNR) of 91.99%, an accuracy of 90.18%, and True Positive Rate (TPR) of 91.08%.

PMID:40243150 | DOI:10.1080/0954898X.2025.2486206

Phys Imaging Radiat Oncol. 2025 Mar 26;34:100749. doi: 10.1016/j.phro.2025.100749. eCollection 2025 Apr.

ABSTRACT

BACKGROUND AND PURPOSE: In order to optimize the radiotherapy treatment and minimize toxicities, organs-at-risk (OARs) and clinical target volume (CTV) must be segmented. Deep Learning (DL) techniques show significant potential for performing this task effectively. The availability of a large single-institute data sample, combined with additional numerous multi-centric data, makes it possible to develop and validate a reliable CTV segmentation model.

MATERIALS AND METHODS: Planning CT data of 1822 patients were available (861 from a single center for training and 961 from 8 centers for validation). A preprocessing step, aimed at standardizing all the images, followed by a 3D-Unet capable of segmenting both right and left CTVs was implemented. The metrics used to evaluate the performance were the Dice similarity coefficient (DSC), the Hausdorff distance (HD), and its 95th percentile variant (HD_95) and the Average Surface Distance (ASD).

RESULTS: The segmentation model achieved high performance on the validation set (DSC: 0.90; HD: 20.5 mm; HD_95: 10.0 mm; ASD: 2.1 mm; epoch 298). Furthermore, the model predicted smoother contours than the clinical ones along the cranial-caudal axis in both directions. When applied to internal and external data the same metrics demonstrated an overall agreement and model transferability for all but one (Inst 9) center.

CONCLUSION: . A 3D-Unet for CTV segmentation trained on a large single institute cohort consisting of planning CTs and manual segmentations was built and externally validated, reaching high performance.

PMID:40242807 | PMC:PMC12002654 | DOI:10.1016/j.phro.2025.100749

KDD. 2025 Aug;2025(v1):2779-2790. doi: 10.1145/3690624.3709402. Epub 2025 Jul 20.

ABSTRACT

Sepsis is an organ dysfunction caused by a deregulated immune response to an infection. Early sepsis prediction and identification allow for timely intervention, leading to improved clinical outcomes. Clinical calculators (e.g., the six-organ dysfunction assessment of SOFA in Figure 1) play a vital role in sepsis identification within clinicians' workflow, providing evidence-based risk assessments essential for sepsis diagnosis. However, artificial intelligence (AI) sepsis prediction models typically generate a single sepsis risk score without incorporating clinical calculators for assessing organ dysfunctions, making the models less convincing and transparent to clinicians. To bridge the gap, we propose to mimic clinicians' workflow with a novel framework SepsisCalc to integrate clinical calculators into the predictive model, yielding a clinically transparent and precise model for utilization in clinical settings. Practically, clinical calculators usually combine information from multiple component variables in Electronic Health Records (EHR), and might not be applicable when the variables are (partially) missing. We mitigate this issue by representing EHRs as temporal graphs and integrating a learning module to dynamically add the accurately estimated calculator to the graphs. Experimental results on real-world datasets show that the proposed model outperforms state-of-the-art methods on sepsis prediction tasks. Moreover, we developed a system to identify organ dysfunctions and potential sepsis risks, providing a human-AI interaction tool for deployment, which can help clinicians understand the prediction outputs and prepare timely interventions for the corresponding dysfunctions, paving the way for actionable clinical decision-making support for early intervention.

PMID:40242786 | PMC:PMC11998859 | DOI:10.1145/3690624.3709402

Exp Ther Med. 2025 Apr 4;29(6):110. doi: 10.3892/etm.2025.12860. eCollection 2025 Jun.

ABSTRACT

TAM (TYRO3, AXL, MERTK) receptor tyrosine kinases (RTKs) have intrinsic roles in tumor cell proliferation, migration, chemoresistance, and suppression of antitumor immunity. The overexpression of TAM RTKs is associated with poor prognosis in various types of cancer. Single-target agents of TAM RTKs have limited efficacy because of an adaptive feedback mechanism resulting from the cooperation of TAM family members. This suggests that multiple targeting of members has the potential for a more potent anticancer effect. The present study used a deep-learning based drug-target interaction (DTI) prediction model called molecule transformer-DTI (MT-DTI) to identify commercially available drugs that may inhibit the three members of TAM RTKs. The results showed that fostamatinib, a spleen tyrosine kinase (Syk) inhibitor, could inhibit the three receptor kinases of the TAM family with an IC50 <1 µM. Notably, no other Syk inhibitors were predicted by the MT-DTI model. To verify this result, this study performed in vitro studies with various types of cancer cell lines. Consistent with the DTI results, this study observed that fostamatinib suppressed cell proliferation by inhibiting TAM RTKs, while other Syk inhibitors showed no inhibitory activity. These results suggest that fostamatinib could exhibit anticancer activity as a pan-TAM inhibitor. Taken together, these findings demonstrated that this artificial intelligence model could be effectively used for drug repurposing and repositioning. Furthermore, by identifying its novel mechanism of action, this study confirmed the potential for fostamatinib to expand its indications as a TAM inhibitor.

PMID:40242601 | PMC:PMC12001310 | DOI:10.3892/etm.2025.12860

Fundam Res. 2023 Jun 18;5(2):880-889. doi: 10.1016/j.fmre.2023.04.018. eCollection 2025 Mar.

ABSTRACT

Mammography is the mainstream imaging modality used for breast cancer screening. Identification of microcalcifications associated with malignancy may result in early diagnosis of breast cancer and aid in reducing the morbidity and mortality associated with the disease. Computer-aided diagnosis (CAD) is a promising technique due to its efficiency and accuracy. Here, we demonstrated that an automated deep-learning pipeline for microcalcification detection and classification on mammography can facilitate early diagnosis of breast cancer. This technique can not only provide the classification results of mammography, but also annotate specific calcification regions. A large mammography dataset was collected, including 4,810 mammograms with 6,663 microcalcification lesions based on biopsy results, of which 3,301 were malignant and 3,362 were benign. The system was developed and tested using images from multiple centers. The overall classification accuracy values for discriminating between benign and malignant breasts were 0.8124 for the training set and 0.7237 for the test set. The sensitivity values of malignant breast cancer prediction were 0.8891 for the training set and 0.7778 for the test set. In addition, we collected information regarding pathological sub-type (pathotype) and estrogen receptor (ER) status, and we subsequently explored the effectiveness of deep learning-based pathotype and ER classification. Automated artificial intelligence (AI) systems may assist clinicians in making judgments and improve their efficiency in breast cancer screening, diagnosis, and treatment.

PMID:40242534 | PMC:PMC11997558 | DOI:10.1016/j.fmre.2023.04.018

Inf Fusion. 2025 Aug;120:103079. doi: 10.1016/j.inffus.2025.103079. Epub 2025 Mar 20.

ABSTRACT

Multivariate time series classification (TSC) is critical for various applications in fields such as healthcare and finance. While various approaches for TSC have been explored, important properties of time series, such as shift equivariance and inversion invariance, are largely underexplored by existing works. To fill this gap, we propose a novel multi-view approach to capture patterns with properties like shift equivariance. Our method integrates diverse features, including spectral, temporal, local, and global features, to obtain rich, complementary contexts for TSC. We use continuous wavelet transform to capture time-frequency features that remain consistent even when the input is shifted in time. These features are fused with temporal convolutional or multilayer perceptron features to provide complex local and global contextual information. We utilize the Mamba state space model for efficient and scalable sequence modeling and capturing long-range dependencies in time series. Moreover, we introduce a new scanning scheme for Mamba, called tango scanning, to effectively model sequence relationships and leverage inversion invariance, thereby enhancing our model's generalization and robustness. Experiments on two sets of benchmark datasets (10+20 datasets) demonstrate our approach's effectiveness, achieving average accuracy improvements of 4.01-6.45% and 7.93% respectively, over leading TSC models such as TimesNet and TSLANet. The code is available at: https://drive.google.com/file/d/1fScmALgreb_sE9_P2kIsQCmt9SNxp7GP/view?usp=sharing.

PMID:40242510 | PMC:PMC11997873 | DOI:10.1016/j.inffus.2025.103079

Rep Pract Oncol Radiother. 2025 Mar 21;30(1):122-125. doi: 10.5603/rpor.104144. eCollection 2025.

NO ABSTRACT

PMID:40242416 | PMC:PMC11999009 | DOI:10.5603/rpor.104144

ACS ES T Water. 2025 Mar 24;5(4):1814-1820. doi: 10.1021/acsestwater.4c01186. eCollection 2025 Apr 11.

ABSTRACT

We present a novel approach for imaging diatoms using lensless imaging and deep learning. We used a laser beam to scatter off samples of diatomaceous earth (diatoms) and then recorded and transformed the scattered light into microscopy images of the diatoms. The predicted microscopy images gave an average SSIM of 0.98 and an average RMSE of 3.26 as compared to the experimental data. We also demonstrate the capability of determining the velocity and angle of movement of the diatoms from their scattering patterns as they were translated through the laser beam. This work shows the potential for imaging and identifying the movement of diatoms and other microsized organisms in situ within the marine environment. Implementing such a method for real-time image acquisition and analysis could enhance environmental management, including improving the early detection of harmful algal blooms.

PMID:40242343 | PMC:PMC11997998 | DOI:10.1021/acsestwater.4c01186

Front Artif Intell. 2025 Apr 2;8:1559461. doi: 10.3389/frai.2025.1559461. eCollection 2025.

ABSTRACT

Adeno-associated virus (AAV) vectors have emerged as powerful tools in gene therapy, potentially treating various genetic disorders. Engineering the AAV capsids through computational methods enables the customization of these vectors to enhance their effectiveness and safety. This engineering allows for the development of gene therapies that are not only more efficient but also personalized to unique genetic profiles. When developing, it is essential to understand the structural biology and the vast techniques used to guide vector designs. This review covers the fundamental biology of the Parvoviridae capsids, focusing on modern structural study techniques, including (a) Cryo-electron microscopy and X-ray Crystallography studies and (b) Comparative analysis of capsid structures across different Parvoviridae species. Along with the structure and evolution of the Parvoviridae capsids, computational methods have provided significant insights into the design of novel AAV vector techniques, which include (a) Structure-guided design of AAV capsids with improved properties, (b) Directed Evolution of AAV capsids for specific applications, and (c) Computational prediction of AAV capsid-receptor interactions. Further discussion addressed the ongoing challenges in the AAV vector design and proposed future directions for exploring enhanced computational tools, such as artificial intelligence/machine learning and deep learning.

PMID:40242328 | PMC:PMC12000042 | DOI:10.3389/frai.2025.1559461

Comput Struct Biotechnol J. 2025 Apr 3;27:1416-1430. doi: 10.1016/j.csbj.2025.04.005. eCollection 2025.

ABSTRACT

The accurate prediction of protein structures remains a cornerstone challenge in structural bioinformatics, essential for understanding the intricate relationship between protein sequence, structure, and function. Recent advancements in Machine Learning (ML) and Deep Learning (DL) have revolutionized this field, offering innovative approaches to tackle one- dimensional (1D) protein structure annotations, including secondary structure, solvent accessibility, and intrinsic disorder. This review highlights the evolution of predictive methodologies, from early machine learning models to sophisticated deep learning frameworks that integrate sequence embeddings and pretrained language models. Key advancements, such as AlphaFold's transformative impact on structure prediction and the rise of protein language models (PLMs), have enabled unprecedented accuracy in capturing sequence-structure relationships. Furthermore, we explore the role of specialized datasets, benchmarking competitions, and multimodal integration in shaping state-of-the-art prediction models. By addressing challenges in data quality, scalability, interpretability, and task-specific optimization, this review underscores the transformative impact of ML, DL, and PLMs on 1D protein prediction while providing insights into emerging trends and future directions in this rapidly evolving field.

PMID:40242292 | PMC:PMC12002955 | DOI:10.1016/j.csbj.2025.04.005

Front Oncol. 2025 Apr 2;15:1557486. doi: 10.3389/fonc.2025.1557486. eCollection 2025.

ABSTRACT

BACKGROUND: Cervical cancer remains a significant global health concern, particularly for women. Accurate preoperative staging is crucial for treatment planning and long-term prognosis. Traditional staging methods rely on manual imaging analysis, which is subjective and time-consuming. Deep learning-based automated staging models offer a promising approach to enhance both accuracy and efficiency.

METHODS: This study retrospectively analyzed preoperative MRI scans (T1 and T2 stages) from 112 cervical cancer patients. Seven deep learning models-DenseNet, FBNet, HRNet, RegNet, ResNet50, ShuffleNet, and ViT-were trained and validated using standardized preprocessing, data augmentation, and manual annotation techniques. Convolutional neural networks were employed to extract multidimensional imaging features, forming the basis of an automated staging prediction model.

RESULTS: Among all tested models, HRNet demonstrated the best performance, achieving an accuracy of 69.70%, recall of 68.89%, F1-score of 68.98%, and precision of 69.62%. ShuffleNet ranked second, with slightly lower performance, while ViT exhibited the weakest predictive ability. The ROC curve analysis confirmed HRNet's superior classification capability, with an AUC of 0.7778, highlighting its effectiveness in small-sample datasets.

CONCLUSION: This study confirms that deep learning models utilizing MRI images can enable automated cervical cancer staging with improved accuracy and efficiency. HRNet, in particular, demonstrates strong potential as a clinical decision-support tool, contributing to the advancement of precision medicine and personalized treatment strategies for cervical cancer.

PMID:40242247 | PMC:PMC11999846 | DOI:10.3389/fonc.2025.1557486

J Nanobiotechnology. 2025 Apr 16;23(1):295. doi: 10.1186/s12951-025-03339-5.

ABSTRACT

BACKGROUND: Lymphoma is a malignant tumor of the immune system and its incidence is increasing year after year, causing a major threat to people's health. Conventional diagnosis of lymphoma basically depends on histological images consuming long-time and tedious manipulations (e.g., 7-15 days) and large-field view (e.g., > 1000 × 1000 μm2). Artificial intelligence has recently revolutionized cancer diagnosis by training pathological image databases via deep learning. Current approaches, however, remain dependent on analyzing wide-field pathological images to detect distinct nuclear, cytologic, and histomorphologic traits for diagnostic categorization, limiting their applicability to minimally invasive lesion.

RESULTS: Herein, we develop a molecular imaging strategy for minimally invasive lymphoma diagnosis. By spreading lymphoma tissue sections tightly on a surface-enhanced Raman scattering (SERS) chip, label-free images of DNA double strand breaks (DSBs) in 30 × 30 μm2 tissue sections could be achieved in ~ 15 min. To establish a proof of concept, the Raman image datasets collected from clinical samples of normal lymphatic tissues and non-Hodgkin's lymphoma (NHL) tissues were well organized and trained in a deep convolutional neural network model, finally achieving a recognition rate of ~ 91.7 ± 2.1%.

CONCLUSIONS: The molecular imaging strategy for minimally invasive lymphoma diagnosis that can achieve a recognition rate of ~ 91.7 ± 2.1%. We anticipate that these results will catalyze the development of a series of histological SERS-AI technologies for diagnosing various diseases, including other types of cancer. In this work, we present a reliable tool to facilitate clinicians in the diagnosis of lymphoma.

PMID:40241186 | DOI:10.1186/s12951-025-03339-5

J Neuroeng Rehabil. 2025 Apr 16;22(1):84. doi: 10.1186/s12984-025-01625-9.

ABSTRACT

Stroke is a serious cerebrovascular disease, and rehabilitation following the acute phase is particularly crucial. Not all rehabilitation outcomes are favorable, highlighting the necessity for personalized rehabilitation. Precision assessment is essential for tailored rehabilitation interventions. Wearable inertial measurement units (IMUs) and deep learning approaches have been effectively employed for motor function prediction. This study aims to use machine learning techniques and data collected from IMUs to assess the Fugl-Meyer upper extremity subscale for post-stroke patients with motor dysfunction. IMUs signals from 120 patients were collected during a clinical trial. These signals were fed into a gated recurrent unit network to complete the scoring of individual actions, which were then aggregated to obtain the total score. Simultaneously, on the basis of the internal correlation between the Fugl-Meyer assessment and the Brunnstrom scale, Brunnstrom stage prediction models of the arm and hand were established via the random forest and extremely randomized trees algorithm. The experimental results show that the proposed models can score Fugl-Meyer items with a high accuracy of 92.66%. The R2 between the doctors' score and the model's score is 0.9838. The Brunnstrom stage prediction models can predict high-quality stages, achieving a Spearman correlation coefficient of 0.9709. The application of the proposed method enables precision assessment of patients' upper extremity motor function, thereby facilitating more personalized rehabilitation programs to achieve optimal recovery outcomes. Trial registration: Clinical trial of telerehabilitation training and intelligent evaluation system, ChiCTR2200061310, Registered 20 June 2022-Retrospective registration.

PMID:40241161 | DOI:10.1186/s12984-025-01625-9

BioData Min. 2025 Apr 16;18(1):31. doi: 10.1186/s13040-025-00445-w.

ABSTRACT

BACKGROUND: Functional deterioration (FD) of various organ systems is the major cause of death in ICU patients, but few studies propose effective multi-task (MT) model to predict FD of multiple organs simultaneously. This study propose a MT deep learning model named inter-organ correlation based multi-task model (IOC-MT), to dynamically predict FD in six organ systems.

METHODS: Three public ICU databases were used for model training and validation. The IOC-MT was designed based on the routine MT deep learning framework, but it used a Graph Attention Networks (GAT) module to capture inter-organ correlation and an adaptive adjustment mechanism (AAM) to adjust prediction. We compared the IOC-MT to five single-task (ST) baseline models, including three deep models (LSTM-ST, GRU-ST, Transformer-ST) and two machine learning models (GRU-ST, RF-ST), and performed ablation study to assess the contribution of important components in IOC-MT. Model discrimination was evaluated by AUROC and AUPRC, and model calibration was assessed by the calibration curve. The attention weight and adjustment coefficient were analyzed at both overall and individual level to show the AAM of IOC-MT.

RESULTS: The IOC-MT had comparable discrimination and calibration to LSTM-ST, GRU-ST and Transformer-ST for most organs under different gap windows in the internal and external validation, and obviously outperformed GRU-ST, RF-ST. The ablation study showed that the GAT, AAM and missing indicator could improve the overall performance of the model. Furthermore, the inter-organ correlation and prediction adjustment of IOC-MT were intuitive and comprehensible, and also had biological plausibility.

CONCLUSIONS: The IOC-MT is a promising MT model for dynamically predicting FD in six organ systems. It can capture inter-organ correlation and adjust the prediction for one organ based on aggregated information from the other organs.

PMID:40241105 | DOI:10.1186/s13040-025-00445-w

BMC Musculoskelet Disord. 2025 Apr 17;26(1):378. doi: 10.1186/s12891-025-08631-x.

ABSTRACT

RATIONALE AND OBJECTIVES: To establish an automated osteoporosis detection model based on low-dose abdominal CT (LDCT). This model combined a deep learning-based automatic segmentation of the proximal femur with a radiomics-based bone status classification.

MATERIALS AND METHODS: A total of 456 participants were retrospectively included and were divided into a development cohort comprising 355 patients, with a 7:3 ratio randomly assigned to the training and validation cohorts, and a test cohort comprising 101 patients. The automatic segmentation model for the proximal femur was trained using VB-Net. The Dice similarity coefficient (DSC) and volume difference (VD) were employed to evaluate the performance of the segmentation model. A three-classification predictive model for assessing bone mineral status was constructed utilizing radiomic analysis. The diagnostic performance of the radiomics model was assessed using the area under the curve (AUC), sensitivity, and specificity.

RESULTS: The automatic segmentation model for the proximal femur demonstrated excellent performance, achieving DSC values of 0.975 ± 0.012 and 0.955 ± 0.137 in the validation and test cohorts, respectively. In the test cohort, the radiomics model utilizing the random forest (RF) classifier achieved AUC values, sensitivity, and specificity of 0.924 (95% CI: 0.854-0.967), 0.846 (95% CI: 0.719-0.931), and 0.837 (95% CI: 0.703-0.927) for the identification of normal bone mass. For the identification of osteoporosis, the corresponding metrics were 0.960 (95% CI: 0.913-1.000), 0.947 (95% CI: 0.740-0.999), and 0.963 (95% CI: 0.897-0.992). In the case of osteopenia, the corresponding metrics were 0.828 (95% CI: 0.747-0.909), 0.767 (95% CI: 0.577-0.901), and 0.746 (95% CI: 0.629-0.842).

CONCLUSION: A three-classification predictive model combining a deep learning-based automatic segmentation of the proximal femur and a radiomics-based bone status classification on LDCT images can be used for the opportunistic detection of osteoporosis.

PMID:40241032 | DOI:10.1186/s12891-025-08631-x

Sci Rep. 2025 Apr 16;15(1):13074. doi: 10.1038/s41598-025-97842-2.

ABSTRACT

Deep learning-based image processing methods are commonly used for bridge crack detection. Aiming at the problem of missed detections and false positives caused by light, stains, and dense cracks during detection, this paper proposes a bridge crack detection algorithm based on the improved YOLOv8n model. Firstly, enhancing the model's feature extraction capabilities by incorporating the global attention mechanism into the Backbone and Neck to gather additional crack characterization information. And optimizing the original feature fusion model through Gam-Concat to enhance the feature fusion effect. Subsequently, in the FPN-PAN structure, replacing the original upsample module with DySample promotes the full fusion of high- and low-resolution feature information, enhancing the detection capability for cracks of different scales. Finally, adding MPDIoU to the Head to optimize the bounding box function loss, enhancing the model's ability to evaluate the overlap of dense cracks and better reflecting the spatial relationships between the cracks. In ablation and comparison experiments, the improved model achieved increases of 3.02%, 3.39%, 2.26%, and 0.81% in mAP@0.5, mAP@0.5:0.95, precision, and recall, respectively, compared to the original model. And the detection accuracy is significantly higher than other comparative models. It has practical application value in bridge inspection projects.

PMID:40240806 | DOI:10.1038/s41598-025-97842-2

Sci Rep. 2025 Apr 16;15(1):13075. doi: 10.1038/s41598-025-97407-3.

ABSTRACT

Multi-Criteria Recommender Systems (MCRSs) improve personalization by incorporating multiple user preferences. However, their application in Technology-Enhanced Learning (TEL) remains limited due to challenges such as data sparsity, over-specialization, and cold-start problems. Traditional techniques, such as Singular Value Decomposition (SVD) and SVD + + , struggle to effectively model the complex interactions within multi-criteria rating data, leading to suboptimal recommendations. This paper introduces a hybrid DeepFM-SVD + + model, which integrates deep learning and factorization-based techniques to improve multi-criteria recommendations. The model captures both low-order feature interactions using factorization machines and high-order dependencies through deep neural networks, enabling more adaptive recommendations. To evaluate its performance, the model is tested on two multi-criteria datasets: ITM-Rec (TEL domain) and Yahoo Movies (non-TEL domain). The experimental results show that DeepFM-SVD + + consistently outperforms the traditional techniques across multiple evaluation metrics. The findings highlight significant improvements in accuracy, demonstrating the model's effectiveness in sparse datasets and its generalization across domains. By addressing the limitations of existing MCRS techniques, this study contributes to advancing personalized learning recommendations in TEL and expands the applicability of deep learning-based MCRS beyond educational contexts.

PMID:40240805 | DOI:10.1038/s41598-025-97407-3

Sci Rep. 2025 Apr 16;15(1):13171. doi: 10.1038/s41598-025-96035-1.

ABSTRACT

X-ray image-based prohibited item detection plays a crucial role in modern public security systems. Despite significant advancements in deep learning, challenges such as feature extraction, object occlusion, and model complexity remain. Although recent efforts have utilized larger-scale CNNs or ViT-based architectures to enhance accuracy, these approaches incur substantial trade-offs, including prohibitive computational overhead and practical deployment limitations. To address these issues, we propose Xray-YOLO-Mamba, a lightweight model that integrates the YOLO and Mamba architectures. Key innovations include the CResVSS block, which enhances receptive fields and feature representation; the SDConv downsampling block, which minimizes information loss during feature transformation; and the Dysample upsampling block, which improves resolution recovery during reconstruction. Experimental results demonstrate that the proposed model achieves superior performance across three datasets, exhibiting robust performance and excellent generalization ability. Specifically, our model attains mAP50-95 of 74.6% (CLCXray), 43.9% (OPIXray), and 73.9% (SIXray), while demonstrating lightweight efficiency with 4.3 M parameters and 10.3 GFLOPs. The architecture achieves real-time performance at 95.2 FPS on the GPUs. In summary, Xray-YOLO-Mamba strikes a favorable balance between precision and computational efficiency, demonstrating significant advantages.

PMID:40240781 | DOI:10.1038/s41598-025-96035-1

Neuroinformatics. 2025 Apr 16;23(2):28. doi: 10.1007/s12021-025-09724-7.

ABSTRACT

Brain slicing is a commonly used technique in brain science research. In order to study the spatial distribution of labeled information, such as specific types of neurons and neuronal circuits, it is necessary to register the brain slice images to the 3D standard brain space defined by the reference atlas. However, the registration of 2D brain slice images to a 3D reference brain atlas still faces challenges in terms of accuracy, computational throughput, and applicability. In this paper, we propose the SlicesMapi, an interactive 3D registration method for brain slice sequence. This method corrects linear and non-linear deformations in both 3D and 2D spaces by employing dual constraints from neighboring slices and corresponding reference atlas slices and guarantees precision by registering images with full resolution, which avoids the information loss of image down-sampling implemented in the deep learning based registration methods. This method was applied to deal the challenges of unknown slice angle registration and non-linear deformations between the 3D Allen Reference Atlas and slices with cytoarchitectonic or autofluorescence channels. Experimental results demonstrate Dice scores of 0.9 in major brain regions, highlighting significant advantages over existing methods. Compared with existing methods, our proposed method is expected to provide a more accurate, robust, and efficient spatial localization scheme for brain slices. Therefore, the proposed method is capable of achieving enhanced accuracy in slice image spatial positioning.

PMID:40240690 | DOI:10.1007/s12021-025-09724-7

Doc Ophthalmol. 2025 Apr 16. doi: 10.1007/s10633-025-10019-0. Online ahead of print.

ABSTRACT

PURPOSE: The electroretinogram (ERG) records the functional response of the retina. In some neurological conditions, the ERG waveform may be altered and could support biomarker discovery. In heterogeneous or rare populations, where either large data sets or the availability of data may be a challenge, synthetic signals with Artificial Intelligence (AI) may help to mitigate against these factors to support classification models.

METHODS: This approach was tested using a publicly available dataset of real ERGs, n = 560 (ASD) and n = 498 (Control) recorded at 9 different flash strengths from n = 18 ASD (mean age 12.2 ± 2.7 years) and n = 31 Controls (mean age 11.8 ± 3.3 years) that were augmented with synthetic waveforms, generated through a Conditional Generative Adversarial Network. Two deep learning models were used to classify the groups using either the real only or combined real and synthetic ERGs. One was a Time Series Transformer (with waveforms in their original form) and the second was a Visual Transformer model utilizing images of the wavelets derived from a Continuous Wavelet Transform of the ERGs. Model performance at classifying the groups was evaluated with Balanced Accuracy (BA) as the main outcome measure.

RESULTS: The BA improved from 0.756 to 0.879 when synthetic ERGs were included across all recordings for the training of the Time Series Transformer. This model also achieved the best performance with a BA of 0.89 using real and synthetic waveforms from a single flash strength of 0.95 log cd s m-2.

CONCLUSIONS: The improved performance of the deep learning models with synthetic waveforms supports the application of AI to improve group classification with ERG recordings.

PMID:40240677 | DOI:10.1007/s10633-025-10019-0