Sci Rep. 2025 Apr 1;15(1):11170. doi: 10.1038/s41598-025-95752-x.

ABSTRACT

Counting the number of people in a room is crucial for optimizing smart buildings, enhancing energy efficiency, and ensuring security while preserving privacy. This study introduces a novel radar-based occupancy estimation method leveraging a 24-GHz Continuous Wave (CW) radar system integrated with time-frequency mapping techniques using Continuous Wavelet Transform (CWT) and power spectrum analysis. Unlike previous studies that rely on WiFi or PIR-based sensors, this approach provides a robust alternative without privacy concerns. The time-frequency scalograms generated from radar echoes were used to train deep-learning models, including DarkNet19, MobileNetV2, and ResNet18. Experiments conducted with sedentary occupants over 4 hours and 40 minutes resulted in 1680 image samples. The proposed approach achieved high accuracy, with DarkNet19 performing the best, reaching 92.7% on CWT images and 92.3% on power spectrum images. Additionally, experiments in a walking environment with another continuous 1 hour of data achieved 86.5% accuracy, demonstrating the method's effectiveness beyond static scenarios. These results confirm that CW radar with deep learning can enable non-intrusive, privacy-preserving occupancy estimation for smart building applications.

PMID:40169921 | DOI:10.1038/s41598-025-95752-x

Sci Rep. 2025 Apr 1;15(1):11076. doi: 10.1038/s41598-025-95672-w.

ABSTRACT

Anti-money laundering has been an issue in our society from the beginning of time. It simply refers to certain regulations and laws set by the government to uncover illegal money, which is passed as legal income. Now, with the emergence of cryptocurrency, it ensures pseudonymity for users. Cryptocurrency is a type of currency that is not authorized by the government and does not exist physically but only on paper. This provides a better platform for criminals for their illicit transactions. New algorithms have been proposed to detect illicit transactions. Machine learning and deep learning algorithms give us hope in identifying these anomalies in transactions. We have selected the Elliptic Bitcoin Dataset. This data set is a graph data set generated from an anonymous blockchain. Each transaction is mapped to real entities with two categories: licit and illicit. Some of them are not labeled. We have run different algorithms for predicting illicit transactions like Logistic Regression, Long Short Term Memory, Support Vector Machine, Random Forest, and a variation of Graph Neural Networks, which is called Graph Convolution Network (GCN). GCN is of special interest in our case. Different evaluation parameters such as accuracy, ROC and F1 score are analyzed for different models. Our experimental results show that the proposed GCN model gives the accuracy [Formula: see text], the AUC 0.9444 and the RMSE 0.1123, which concludes that our GCN is better than the existing models, in particular with the model proposed in Weber et al. (Anti-money laundering in bitcoin: experimenting with graph convolutional networks for financial forensics, 2019. http://arxiv.org/abs/1908.02591 ).

PMID:40169862 | DOI:10.1038/s41598-025-95672-w

Sci Rep. 2025 Apr 1;15(1):11058. doi: 10.1038/s41598-025-95381-4.

ABSTRACT

The fast development of social media platforms has led to an unprecedented growth of daily short text content. Removing valued patterns and insights from this vast amount of textual data requires advanced methods to provide information while preserving its essential components successfully. A text summarization system takes more than one document as input and tries to give a fluent and concise summary of the most significant information in the input. Recent solutions for condensing and reading text are ineffective and time-consuming, provided plenty of information is available online. Concerning this challenge, automated text summarization methods have developed as a convincing choice, achieving important significance in their growth. It was separated into two kinds according to the abstraction methods utilized: abstractive summarization (AS) and extractive summarization (ES). Furthermore, automatic text summarization has many applications and spheres of impact. This manuscript proposes an Adaptive Search Mechanism Based Hierarchical Learning Networks for Social Media Data Summarization and Classification Model (ASMHLN-SMDSCM) technique. The ASMHLN-SMDSCM approach aims to present a novel approach for text summarization on social media using advanced deep learning models. To accomplish that, the proposed ASMHLN-SMDSCM model performs text pre-processing, which contains dissimilar levels employed to handle unprocessed data. The BERT model is used for the feature extraction process. Furthermore, the moth search algorithm (MSA)-based hyperparameter selection process is performed to optimize the feature extraction results of the BERT model. Finally, the classification uses the TabNet and convolutional neural network (TabNet + CNN) model. The efficiency of the ASMHLN-SMDSCM method is validated by comprehensive studies using the FIFA and FARMER datasets. The experimental validation of the ASMHLN-SMDSCM method illustrated a superior accuracy value of 98.87% and 98.55% over recent techniques.

PMID:40169845 | DOI:10.1038/s41598-025-95381-4

Sci Rep. 2025 Apr 1;15(1):11053. doi: 10.1038/s41598-025-93536-x.

ABSTRACT

This paper presents an enhanced ensemble classification framework for Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) under diverse operational conditions, including Standard Operating Conditions (SOC) and Extended Operating Conditions (EOC). The proposed method integrates the strengths of Residual Neural Networks (ResNet) replacing Convolutional Neural Networks (CNN), Support Vector Machines (SVM), and template matching, leveraging majority voting to combine their complementary capabilities. The ensemble framework achieves improved robustness and classification accuracy across varied scenarios. The methodology employs ResNet, a deep learning architecture known for its superior feature extraction and classification capabilities, replacing AlexNet to address limitations in generalization and consistency. ResNet demonstrated better performance with average accuracies of 92.67% under SOC and 88.9% under EOC, showing consistent results across all six target classes, as compared to the CNN-based ensemble approach with average accuracies of 90.30% under SOC and 87.22% under EOC. The SVM is employed for its robustness in handling overfitting and classifying features extracted from 16 region properties. Template matching is included for its resilience in challenging conditions where deep learning techniques may underperform. Experimental validation using the MSTAR dataset, a standard benchmark for SAR ATR, highlights the effectiveness of this ensemble approach. The results confirm significant improvements in classification accuracy and robustness over individual classifiers, demonstrating the practical applicability of the ensemble approach to real-world SAR ATR challenges. This research advances SAR ATR by addressing critical challenges, including noise, occlusion, and variations in viewing angles while achieving high classification performance under diverse conditions. The integration of ResNet further enhances the framework's adaptability and reliability.

PMID:40169814 | DOI:10.1038/s41598-025-93536-x

Sci Rep. 2025 Apr 1;15(1):11127. doi: 10.1038/s41598-025-93059-5.

ABSTRACT

Bladder (BL) cancer is the 10th most common cancer worldwide, ranking 9th in males and 13th in females in the United States, respectively. BL cancer is a quick-growing tumor of all cancer forms. Given a malignant tumor's high malignancy, rapid metastasis prediction and accurate treatment are critical. The most significant drivers of the intricate genesis of cancer are complex genetics, including deoxyribonucleic acid (DNA) insertions and deletions, abnormal structure, copy number variations (CNVs), and single nucleotide variations (SNVs). The proposed method enhances the identification of driver genes at the individual patient level by employing attention mechanisms to extract features of both coding and non-coding genes and predict BL cancer based on the personalized driver gene (PDG) detection. The embedded vectors are propagated through the three dense blocks for the binary classification of PDGs. The novel constructure of graph neural network (GNN) with attention mechanism, called Multi Stacked-Layered GAT (MSL-GAT) leverages graph attention mechanisms (GAT) to identify and predict critical driver genes associated with BL cancer progression. In order to pick out and extract essential features from both coding and non-coding genes, including long non-coding RNAs (lncRNAs), which are known to be crucial to the advancement of BL cancer. The approach analyzes key genetic changes (such as SNVs, CNVs, and structural abnormalities) that lead to tumorigenesis and metastasis by concentrating on personalized driver genes (PDGs). The discovery of genes crucial for the survival and proliferation of cancer cells is made possible by the model's precise classification of PDGs. MSL-GAT draws attention to certain lncRNAs and other non-coding elements that control carcinogenic pathways by utilizing the attention mechanism. Tumor development, metastasis, and medication resistance are all facilitated by these lncRNAs, which are frequently overexpressed or dysregulated in BL cancer. In order to reduce the survival of cancer cells, the model's predictions can direct specific treatment approaches, such as RNA interference (RNAi), to mute or suppress the expression of these important genes. MSL-GAT is followed by three dense blocks that spread the embedded vectors to categorize PDGs, making it possible to determine which genes are more likely to cause BL cancer in a certain patient. The model facilitates the identification of new treatment targets by offering a thorough understanding of the molecular landscape of BL cancer through the integration of multi-omics data, encompassing as genomic, transcriptomic, and epigenomic metadata. We compared the novel approach with classical machine learning methods and other deep learning-based methods on benchmark TCGA-BLCA, and the leave-one-out experimental results showed that MSL-GAT achieved better performance than competitive methods. This approach achieves accuracy with 97.72% and improves specificity and sensitivity. It can potentially aid physicians during early prediction of BL cancer.

PMID:40169776 | DOI:10.1038/s41598-025-93059-5

J Clin Oncol. 2025 Apr 1:JCO2401501. doi: 10.1200/JCO-24-01501. Online ahead of print.

ABSTRACT

PURPOSE: Artificial intelligence (AI) holds significant promise for improving cancer diagnosis and treatment. Here, we present a foundation AI model for prognosis prediction on the basis of standard hematoxylin and eosin-stained histopathology slides.

METHODS: In this multinational cohort study, we developed AI models to predict prognosis from histopathology images of patients with GI cancers. First, we trained a foundation model using over 130 million patches from 104,876 whole-slide images on the basis of self-supervised learning. Second, we fine-tuned deep learning models for predicting survival outcomes and validated them across seven cohorts, including 1,619 patients with gastric and esophageal cancers and 2,594 patients with colorectal cancer. We further assessed the model for predicting survival benefit from adjuvant chemotherapy.

RESULTS: The AI models predicted disease-free survival and disease-specific survival with a concordance index of 0.726-0.797 for gastric cancer and 0.714-0.757 for colorectal cancer in the validation cohorts. The models stratified patients into high-risk and low-risk groups, with 5-year survival rates of 49%-52% versus 76%-92% in gastric cancer and 43%-72% versus 81%-98% in colorectal cancer. In multivariable analysis, the AI risk scores remained an independent prognostic factor after adjusting for clinicopathologic variables. Compared with stage alone, an integrated model consisting of stage and image information improved prognosis prediction across all validation cohorts. Finally, adjuvant chemotherapy was associated with improved survival in the high-risk group but not in the low-risk group (treatment-model interaction P = .01 and .006) for stage II/III gastric and colorectal cancer, respectively.

CONCLUSION: The pathology foundation model can accurately predict survival outcomes and complement clinicopathologic factors in GI cancers. Pending prospective validation, it may be used to improve risk stratification and inform personalized adjuvant therapy.

PMID:40168636 | DOI:10.1200/JCO-24-01501

PLoS One. 2025 Apr 1;20(4):e0321410. doi: 10.1371/journal.pone.0321410. eCollection 2025.

ABSTRACT

[This corrects the article DOI: 10.1371/journal.pone.0312177.].

PMID:40168295 | DOI:10.1371/journal.pone.0321410

IEEE Trans Cybern. 2025 Apr 1;PP. doi: 10.1109/TCYB.2025.3539990. Online ahead of print.

ABSTRACT

As a longstanding scientific challenge, accurate and timely ocean forecasting has always been a sought-after goal for ocean scientists. However, traditional theory-driven numerical ocean prediction (NOP) suffers from various challenges, such as the indistinct representation of physical processes, inadequate application of observation assimilation, and inaccurate parameterization of models, which lead to difficulties in obtaining effective knowledge from massive observations, and enormous computational challenges. With the successful evolution of data-driven deep learning in various domains, it has been demonstrated to mine patterns and deep insights from the ever-increasing stream of oceanographic spatiotemporal data, which provides novel possibilities for revolution in ocean forecasting. Deep-learning-based ocean forecasting (DLOF) is anticipated to be a powerful complement to NOP. Nowadays, researchers attempt to introduce deep learning into ocean forecasting and have achieved significant progress that provides novel motivations for ocean science. This article provides a comprehensive review of the state-of-the-art DLOF research regarding model architectures, spatiotemporal multiscales, and interpretability while specifically demonstrating the feasibility of developing hybrid architectures that incorporate theory-driven and data-driven models. Moreover, we comprehensively evaluate DLOF from datasets, benchmarks, and cloud computing. Finally, the limitations of current research and future trends of DLOF are also discussed and prospected.

PMID:40168238 | DOI:10.1109/TCYB.2025.3539990

IEEE Trans Cybern. 2025 Mar 31;PP. doi: 10.1109/TCYB.2025.3550203. Online ahead of print.

ABSTRACT

The acceptance of academic papers involves a complex peer-review process that requires substantial human and material resources and is susceptible to biases. With advancements in deep learning technologies, researchers have explored automated approaches for assessing paper acceptance. Existing automated academic paper rating methods primarily rely on the full content of papers to estimate acceptance probabilities. However, these methods are often inefficient and introduce redundant or irrelevant information. Additionally, while Bert can capture general semantic representations through pretraining on large-scale corpora, its performance on the automatic academic paper rating (AAPR) task remains suboptimal due to discrepancies between its pretraining corpus and academic texts. To address these issues, this study proposes LMCBert, a model that integrates large language models (LLMs) with momentum contrastive learning (MoCo). LMCBert utilizes LLMs to extract the core semantic content of papers, reducing redundancy and improving the understanding of academic texts. Furthermore, it incorporates MoCo to optimize Bert training, enhancing the differentiation of semantic representations and improving the accuracy of paper acceptance predictions. Empirical evaluations demonstrate that LMCBert achieves effective performance on the evaluation dataset, supporting the validity of the proposed approach. The code and data used in this article are publicly available at https://github.com/iioSnail/LMCBert.

PMID:40168236 | DOI:10.1109/TCYB.2025.3550203

IEEE Trans Med Imaging. 2025 Mar 31;PP. doi: 10.1109/TMI.2025.3556310. Online ahead of print.

ABSTRACT

The scarcity of annotations has become a significant obstacle in training powerful deep-learning models for medical image segmentation, limiting their clinical application. To overcome this, semi-supervised learning that leverages abundant unlabeled data is highly desirable to enhance model training. However, most existing works still focus on specific medical tasks and underestimate the potential of learning across diverse tasks and datasets. In this paper, we propose a Versatile Semi-supervised framework (VerSemi) to present a new perspective that integrates various SSL tasks into a unified model with an extensive label space, exploiting more unlabeled data for semi-supervised medical image segmentation. Specifically, we introduce a dynamic task-prompted design to segment various targets from different datasets. Next, this unified model is used to identify the foreground regions from all labeled data, capturing cross-dataset semantics. Particularly, we create a synthetic task with a CutMix strategy to augment foreground targets within the expanded label space. To effectively utilize unlabeled data, we introduce a consistency constraint that aligns aggregated predictions from various tasks with those from the synthetic task, further guiding the model to accurately segment foreground regions during training. We evaluated our VerSemi framework against seven established SSL methods on four public benchmarking datasets. Our results suggest that VerSemi consistently outperforms all competing methods, beating the second-best method with a 2.69% average Dice gain on four datasets and setting a new state of the art for semi-supervised medical image segmentation. Code is available at https://github.com/maxwell0027/VerSemi.

PMID:40168233 | DOI:10.1109/TMI.2025.3556310

IEEE Trans Med Imaging. 2025 Apr 1;PP. doi: 10.1109/TMI.2025.3556420. Online ahead of print.

ABSTRACT

Graph deep learning (GDL) has demonstrated impressive performance in predicting population-based brain disorders (BDs) through the integration of both imaging and non-imaging data. However, the effectiveness of GDL-based methods heavily depends on the quality of modeling multi-modal population graphs and tends to degrade as the graph scale increases. Moreover, these methods often limit interactions between imaging and non-imaging data to node-edge interactions within the graph, overlooking complex inter-modal correlations and resulting in suboptimal outcomes. To address these challenges, we propose MM-GTUNets, an end-to-end Graph Transformer-based multi-modal graph deep learning (MMGDL) framework designed for large-scale brain disorders prediction. To effectively utilize rich multi-modal disease-related information, we introduce Modality Reward Representation Learning (MRRL), which dynamically constructs population graphs using an Affinity Metric Reward System (AMRS). We also employ a variational autoencoder to reconstruct latent representations of non-imaging features aligned with imaging features. Based on this, we introduce Adaptive Cross-Modal Graph Learning (ACMGL), which captures critical modality-specific and modality-shared features through a unified GTUNet encoder, taking advantages of Graph UNet and Graph Transformer, along with a feature fusion module. We validated our method on two public multi-modal datasets ABIDE and ADHD-200, demonstrating its superior performance in diagnosing BDs. Our code is available at https://github.com/NZWANG/MM-GTUNets.

PMID:40168232 | DOI:10.1109/TMI.2025.3556420

IEEE J Biomed Health Inform. 2025 Apr 1;PP. doi: 10.1109/JBHI.2025.3556775. Online ahead of print.

ABSTRACT

Epilepsy affects millions worldwide, posing significant challenges due to the erratic and unexpected nature of seizures. Despite advancements, existing seizure prediction techniques remain limited in their ability to forecast seizures with high accuracy, impacting the quality of life for those with epilepsy. This research introduces the Coherence-based Seizure Prediction (CoSP) method, which integrates coherence analysis with deep learning to enhance seizure prediction efficacy. In CoSP, electroencephalography (EEG) recordings are divided into 10-second segments to extract channel pairwise coherence. This coherence data is then used to train a four-layer convolutional neural network to predict the probability of being in a preictal state. The predicted probabilities are then processed to issue seizure warnings. CoSP was evaluated in a pseudo-prospective setting using long-term iEEG data from ten patients in the NeuroVista seizure advisory system. CoSP demonstrated promising predictive performance across a range of preictal intervals (4 to 180 minutes). CoSP achieved a median Seizure Sensitivity (SS) of 0.79, a median false alarm rate of 0.15 per hour, and a median Time in Warning (TiW) of 27%, highlighting its potential for accurate and reliable seizure prediction. Statistical analysis confirmed that CoSP significantly outperformed chance (p = 0.001) and other baseline methods (p <0.05) under similar evaluation configurations.

PMID:40168220 | DOI:10.1109/JBHI.2025.3556775

IEEE J Biomed Health Inform. 2025 Apr 1;PP. doi: 10.1109/JBHI.2025.3556676. Online ahead of print.

ABSTRACT

Deformable image registration is crucial for medical image analysis, yet the complexity of deep learning networks often limits their deployment on resource-limited devices. Current distillation methods in registration tasks fail to effectively transfer complex deformation handling capabilities to non-iterative lightweight networks, leading to insignificant performance improvement. To address this, we propose the Framework for Iterative to Non-iterative Distillation (FIND), which efficiently transfers these capabilities to a Non-Iterative Lightweight (NIL) network. FIND employs a dual-step process: first, using recurrent distillation to derive a high-performance non-iterative teacher assistant from an iterative network; second, using advanced feature distillation from the assistant to the lightweight network. This enables NIL to perform rapid, effective registration on resource-limited devices. Experiments across four datasets show that NIL can achieve up to 60 times faster performance on CPU and 89 times on GPU than compared deep learning methods, with superior registration accuracy improvements of up to 3.5 points in Dice scores. Code is available at https://anonymous.4open.science/r/FIND-7A16.

PMID:40168217 | DOI:10.1109/JBHI.2025.3556676

IEEE J Biomed Health Inform. 2025 Apr 1;PP. doi: 10.1109/JBHI.2025.3556717. Online ahead of print.

ABSTRACT

Substantial progress has been made in developing deep-learning models for clinical diagnosis. While excelling in diagnostics, the broader clinical decision-making process also involves establishing optimal follow-up intervals (TCU), crucial for prognosis and timely treatment. To fully support clinical practice, it is imperative that deep learning models contribute to both initial diagnosis and TCU prediction. However, relying on separate monolithic models is computationally demanding and lacks interpretability, hindering clinician trust. Our proposed bilateral model, emphasizing ophthalmological cases, offers both initial diagnoses and follow-up predictions, enhancing interpretability and trust in clinical applications as clinicians are more likely to trust recommendations, knowing the diagnosis used is correct. Inspired by clinical practice, the model integrates hierarchical inference and self-supervised learning techniques to enhance predictive accuracy and interpretability. Consisting of a sparse autoencoder, diagnosis classifier, and TCU classifier, the model leverages insights from clinicians and observations of ophthalmological datasets to capture salient features and facilitate robust learning. By employing shared weights for encoding and diagnosing each organ, the model optimizes efficiency and doubles the effective dataset size. Experimental results on an ophthalmological dataset demonstrate superior performance compared to baseline models, with the hierarchical inference structure providing valuable insights into the model's decision-making process. The bilateral model not only enhances predictive modeling for conditions affecting bilaterally symmetrical organs but also empowers clinicians with interpretable outputs crucial for informed clinical decision-making, thereby advancing clinical practice and improving patient care.

PMID:40168215 | DOI:10.1109/JBHI.2025.3556717

IEEE Trans Neural Syst Rehabil Eng. 2025 Apr 1;PP. doi: 10.1109/TNSRE.2025.3555542. Online ahead of print.

ABSTRACT

Auditory attention decoding from electroencephalogram (EEG) could infer to which source the user is attending in noisy environments. Decoding algorithms and experimental paradigm designs are crucial for the development of technology in practical applications. To simulate real-world scenarios, this study proposed a cue-masked auditory attention paradigm to avoid information leakage before the experiment. To obtain high decoding accuracy with low latency, an end-to-end deep learning model, AADNet, was proposed to exploit the spatiotemporal information from the short time window of EEG signals. The results showed that with a 0.5-second EEG window, AADNet achieved an average accuracy of 93.46% and 91.09% in decoding auditory orientation attention (OA) and timbre attention (TA), respectively. It significantly outperformed five previous methods and did not need the knowledge of the original audio source. This work demonstrated that it was possible to detect the orientation and timbre of auditory attention from EEG signals fast and accurately. The results are promising for the real-time multi-property auditory attention decoding, facilitating the application of the neuro-steered hearing aids and other assistive listening devices.

PMID:40168202 | DOI:10.1109/TNSRE.2025.3555542

ACS Sens. 2025 Apr 1. doi: 10.1021/acssensors.4c03621. Online ahead of print.

ABSTRACT

Lung cancer exhibits strong heterogeneity, and its early diagnosis and precise subtyping are of great importance, as they can increase the ability to deliver personalized medicines by tailoring therapy regimens. Tissue biopsy, albeit the gold standard, is invasive, costly and provides limited information about the tumor and its molecular landscape. Exosomes, as promising biomarkers for lung cancer, are a heterogeneous collection of membranous vesicles containing tumor-specific information for liquid biopsy to identify lung cancer subtypes. However, the small size, complex structure, and heterogeneous molecular features of exosomes pose significant challenges for their effective isolation and analysis. Herein, we report a deep learning-driven microfluidic chip with surface-enhanced Raman scattering (SERS) readout to characterize the differences in exosomes for the early diagnosis and molecular subtyping of non-small cell lung cancer (NSCLC). This integration comprises a processing unit for exosome capture and enrichment using polystyrene microspheres (PS) binding gold nanocubes (AuNCs) and anti-CD-9 antibody (denoted as PACD), and an optical sensing unit to trap the PACD and detect SERS signals from these exosomes. This system achieved a maximum trapping efficiency of 85%, and could distinguish three different NSCLC cell lines from the normal cell line with an overall accuracy of 97.88% and an area under the curve (AUC) of over 0.95 for each category. This work highlights the combined power of deep learning, SERS, and microfluidics in realizing the capture, detection, and analysis of exosomes from biological matrices, which may pave the way for clinical exosome-based cancer diagnosis and prognostication in the future.

PMID:40167999 | DOI:10.1021/acssensors.4c03621

Radiol Med. 2025 Apr 1. doi: 10.1007/s11547-025-02006-x. Online ahead of print.

ABSTRACT

PURPOSE: This study aims to evaluate the effectiveness of combining transvaginal ultrasound (US)-based radiomics and deep learning model for the accurate differentiation between benign and malignant ovarian tumors in large-scale studies.

MATERIALS AND METHODS: A multicenter retrospective study collected grayscale and color US images of ovarian tumors. Patients were divided into training, internal, and external validation groups. Models including a convolutional neural networks (CNN), optimal radiomics, and a combined model were constructed and evaluated for predictive performance using area under curve (AUC), sensitivity, and specificity. The DeLong test compared model AUCs with O-RADS and expert assessments.

RESULTS: 3193 images from 2078 patients were analyzed. The CNN achieved AUCs of 0.970 (internal) and 0.959 (external), respectively. Optimal radiomic model achieved AUCs of 0.949 (internal) and 0.954 (external), respectively. The combined CNN-radiomics model attained the highest AUC of 0.977 (internal) and 0.972 (external), respectively, outperforming individual models, O-RADS, and expert methods (p < 0.05).

CONCLUSIONS: The combined CNN-radiomics model using transvaginal US images provides more accurate and reliable ovarian tumor diagnosis, enhancing malignancy prediction and offering clinicians a more precise diagnostic tool.

PMID:40167932 | DOI:10.1007/s11547-025-02006-x

Discov Oncol. 2025 Apr 1;16(1):438. doi: 10.1007/s12672-025-02212-z.

ABSTRACT

Artificial intelligence (AI) marks a frontier in histopathologic analysis shift towards the clinic, becoming a mainstream choice to interpret histological images. Surveying studies assessing AI applications in histopathology from 2013 to 2024, we review key methods (including supervised, unsupervised, weakly supervised and transfer learning) in deep learning-based pattern recognition in computational histopathology for diagnostic and prognostic purposes. Deep learning methods also showed utility in identifying a wide range of genetic mutations and standard pathology biomarkers from routine histology. This survey of 41 primary studies also encompasses key regions of AI applicability in histopathology in a multi-cancer review while marking prospects to introduce AI into the clinical setting with key examples including Swarm Learning and Data Fusion.

PMID:40167870 | DOI:10.1007/s12672-025-02212-z

Opt Lett. 2025 Apr 1;50(7):2390-2393. doi: 10.1364/OL.549356.

ABSTRACT

To meet rapidly growing communication demands, researchers have focused on structured light-based shift keying techniques. However, higher-order modes are prone to large diffraction divergence and are easily perturbed. In this study, we experimentally demonstrate what we believe to be a novel coherence shaping method for petal-like structures of optical vortices, enabling the generation of non-diffraction interference states between completely coherent and incoherent states. In addition, we propose a coherence shift keying (CSK) scheme enabled by deep learning, and a well-trained model can achieve a high recognition accuracy (>0.997) of interference states under practical conditions, including complex environments. Further experimental validation has confirmed that the minimum achievable visibility-level bandwidth is 0.02. This study provides a new, to the best of our knowledge, platform for low-order structured light mode-based high-capacity and encrypted shift keying communication systems.

PMID:40167728 | DOI:10.1364/OL.549356

Abdom Radiol (NY). 2025 Apr 1. doi: 10.1007/s00261-025-04895-y. Online ahead of print.

ABSTRACT

PURPOSE: To investigate the value of deep learning (DL) in T2-weighted imaging (T2DL) of the bladder regarding acquisition time (TA), image quality, and diagnostic confidence compared to standard T2-weighted turbo-spin-echo (TSE) imaging (T2S).

METHODS: We prospectively enrolled a total of 28 consecutive patients for the evaluation of bladder cancer. T2S and T2DL sequences in three planes were performed for each participant, and acquisition time was compared between the two acquisition protocols. The image evaluation was conducted independently by two radiologists using a 5-point Likert scale for artifacts, noise, overall image quality, and diagnostic confidence, with 5 indicating the best quality. Additionally, T2 scoring based on Vesical Imaging-Reporting and Data System (VI-RADS) was performed by two readers.

RESULTS: Compared to T2S, the acquisition time of T2DL was reduced by 49.4% in the axial and by 43.8% in the coronal and sagittal orientations. The severity and impact of artifacts and noise levels were superior in T2DL versus T2S (both p < 0.05). The overall image quality in T2DL was demonstrated to be higher compared to that in T2S in axial and sagittal imaging (both p < 0.05). The diagnostic confidence and T2 scoring of both sequences in all planes did not differ (p > 0.05).

CONCLUSIONS: Our study preliminarily demonstrated the feasibility of T2-weighted imaging with DL reconstruction of bladder MR in clinical practice. T2DL achieved a reduction in acquisition time, superior lesion detectability, and overall image quality with similar diagnostic confidence and T2 score compared to the standard T2 TSE sequence.

PMID:40167648 | DOI:10.1007/s00261-025-04895-y