Prostate Cancer. 2025 May 8;2025:2019841. doi: 10.1155/proc/2019841. eCollection 2025.

ABSTRACT

Introduction: Prostate cancer is the second most common cancer among men worldwide. This cancer has become extremely noticeable due to the increase of prostate cancer in Iranian men in recent years due to the lack of marriage and sexual intercourse, as well as the abuse of hormones in sports without any standards. Methods: The histopathology images from a treatment center to diagnose prostate cancer are used with the help of deep learning methods, considering the two characteristics of Tile and Grad-CAM. The approach of this research is to present a prostate cancer diagnosis model to achieve proper performance from histopathology images with the help of a developed deep learning method based on the manifold model. Results: Similarly, in addition to the diagnosis of prostate cancer, a study on the methods of preventing this disease was investigated in literature reviews, and finally, after simulation, prostate cancer presentation factors were determined. Conclusions: The simulation results indicated that the proposed method has a performance advantage over the other state-of-the-art methods, and the accuracy of this method is up to 97.41%.

PMID:40376132 | PMC:PMC12081159 | DOI:10.1155/proc/2019841

Front Physiol. 2025 May 1;16:1581699. doi: 10.3389/fphys.2025.1581699. eCollection 2025.

ABSTRACT

Freezing of Gait (FoG) is a disabling motor symptom that characterizes Parkinson's Disease (PD) patients and significantly affects their mobility and quality of life. The paper presents a novel hybrid deep learning framework for the detection of FoG episodes using wearable sensors. The methodology combines CNNs for spatial feature extraction, BiLSTM networks for temporal modeling, and an attention mechanism to enhance interpretability and focus on critical gait features. The approach leverages multimodal datasets, including tDCS FOG, DeFOG, Daily Living, and Hantao's Multimodal, to ensure robustness and generalizability. The proposed model deals with sensor noise, inter-subject variability, and data imbalance through comprehensive preprocessing techniques such as sensor fusion, normalization, and data augmentation. The proposed model achieved an average accuracy of 92.5%, F1-score of 89.3%, and AUC of 0.91, outperforming state-of-the-art methods. Post-training quantization and pruning enabled deployment on edge devices such as Raspberry Pi and Coral TPU, achieving inference latency under 350 ms. Ablation studies show the critical contribution of key architectural components to the model's effectiveness. Optimized to be deployed real-time, it is a potentially promising solution that can help correctly detect FoG, thereby achieving better clinical monitoring and improving patients' outcomes in a controlled as well as real world.

PMID:40376117 | PMC:PMC12079673 | DOI:10.3389/fphys.2025.1581699

bioRxiv [Preprint]. 2025 Apr 18:2025.04.13.648574. doi: 10.1101/2025.04.13.648574.

ABSTRACT

Deep neural networks have led to significant advancements in microscopy image generation and analysis. In single-molecule localization based super-resolution microscopy, neural networks are capable of predicting fluorophore positions from high-density emitter data, thus reducing acquisition time, and increasing imaging throughput. However, neural network-based solutions in localization microscopy require intensive human intervention and computation expertise to address the compromise between model performance and its generalization. For example, researchers manually tune parameters to generate training images that are similar to their experimental data; thus, for every change in the experimental conditions, a new training set should be manually tuned, and a new model should be trained. Here, we introduce AutoDS and AutoDS3D, two software programs for reconstruction of single-molecule super-resolution microscopy data that are based on Deep-STORM and DeepSTORM3D, that significantly reduce human intervention from the analysis process by automatically extracting the experimental parameters from the imaging raw data. In the 2D case, AutoDS selects the optimal model for the analysis out of a set of pre-trained models, hence, completely removing user supervision from the process. In the 3D case, we improve the computation efficiency of DeepSTORM3D and integrate the lengthy workflow into a graphic user interface that enables image reconstruction with a single click. Ultimately, we demonstrate superior performance of both pipelines compared to Deep-STORM and DeepSTORM3D for single-molecule imaging data of complex biological samples, while significantly reducing the manual labor and computation time.

PMID:40376092 | PMC:PMC12080944 | DOI:10.1101/2025.04.13.648574

Front Immunol. 2025 May 1;16:1513589. doi: 10.3389/fimmu.2025.1513589. eCollection 2025.

ABSTRACT

INTRODUCTION: Pancreatic ductal adenocarcinoma (PDAC) exhibits higher hypoxia level than most solid tumors, and the presence of intratumoral hypoxia is associated with a poor prognosis. However, the identification of hypoxia levels based on pathological images, and the mechanisms regulating ferroptosis resistance, remain to be elucidated. The objective of this study was to construct a deep learning model to evaluate the hypoxia characteristics of PDAC and to explore the role of Sulfide quinone oxidoreductase (SQOR) in hypoxia-mediated ferroptosis resistance.

METHODS: Multi-omics data were integrated to analyze the correlation between hypoxia score of PDAC, SQOR expression and prognosis, and ferroptosis resistance level. A deep learning model of Whole Slide Images (WSIs) were constructed to predict the hypoxia level of patients. In vitro hypoxia cell models, SQOR knockdown experiments and nude mouse xenograft models were used to verify the regulatory function of SQOR on ferroptosis.

RESULTS: PDAC exhibited significantly higher hypoxia levels than normal tissues, correlating with reduced overall survival in patients. In slide level, our deep learning model can effectively identify PDAC hypoxia levels with good performance. SQOR was upregulated in tumor tissues and positively associated with both hypoxia score and ferroptosis resistance. SQOR promotes the malignant progression of PDAC in hypoxic environment by enhancing the resistance of tumor cells to ferroptosis. SQOR knockdown resulted in decreased cell viability, decreased migration ability and increased MDA level under hypoxic Ersatin induced conditions. Furthermore, SQOR inhibitor in combination with ferroptosis inducer has the potential to inhibit tumor growth in vivo in a synergistic manner.

DISCUSSION: This study has established a hypoxia detection model of PDAC based on WSIs, providing a new tool for clinical evaluation. The study revealed a new mechanism of SQOR mediating ferroptosis resistance under hypoxia and provided a basis for targeted therapy.

PMID:40375994 | PMC:PMC12078260 | DOI:10.3389/fimmu.2025.1513589

Proc SPIE Int Soc Opt Eng. 2024 Feb;12933:1293319. Epub 2024 Apr 3.

ABSTRACT

Segmentation of microvascular structures, such as arterioles, venules, and capillaries, from human kidney whole slide images (WSI) has become a focal point in renal pathology. Current manual segmentation techniques are time-consuming and not feasible for large-scale digital pathology images. While deep learning-based methods offer a solution for automatic segmentation, most suffer from a limitation: they are designed for and restricted to training on single-site, single-scale data. In this paper, we present Omni-Seg, a novel single dynamic network method that capitalizes on multi-site, multi-scale training data. Unique to our approach, we utilize partially labeled images, where only one tissue type is labeled per training image, to segment microvascular structures. We train a singular deep network using images from two datasets, HuBMAP and NEPTUNE, across different magnifications (40×, 20×, 10×, and 5×). Experimental results indicate that Omni-Seg outperforms in terms of both the Dice Similarity Coefficient (DSC) and Intersection over Union (IoU). Our proposed method provides renal pathologists with a powerful computational tool for the quantitative analysis of renal microvascular structures.

PMID:40375952 | PMC:PMC12080525

J Med Internet Res. 2025 May 15;27:e69785. doi: 10.2196/69785.

ABSTRACT

BACKGROUND: Chronic obstructive pulmonary disease (COPD) is a common and progressive respiratory condition characterized by persistent airflow limitation and symptoms such as dyspnea, cough, and sputum production. Acute exacerbations (AE) of COPD (AE-COPD) are key determinants of disease progression; yet, existing predictive models relying mainly on spirometric measurements, such as forced expiratory volume in 1 second, reflect only a fraction of the physiological information embedded in respiratory function tests. Recent advances in artificial intelligence (AI) have enabled more sophisticated analyses of full spirometric curves, including flow-volume loops and volume-time curves, facilitating the identification of complex patterns associated with increased exacerbation risk.

OBJECTIVE: This study aimed to determine whether a predictive model that integrates clinical data and spirometry images with the use of AI improves accuracy in predicting moderate-to-severe and severe AE-COPD events compared to a clinical-only model.

METHODS: A retrospective cohort study was conducted using COPD registry data from 2 teaching hospitals from January 2004 to December 2020. The study included a total of 10,492 COPD cases, divided into a development cohort (6870 cases) and an external validation cohort (3622 cases). The AI-enhanced model (AI-PFT-Clin) used a combination of clinical variables (eg, history of AE-COPD, dyspnea, and inhaled treatments) and spirometry image data (flow-volume loop and volume-time curves). In contrast, the Clin model used only clinical variables. The primary outcomes were moderate-to-severe and severe AE-COPD events within a year of spirometry.

RESULTS: In the external validation cohort, the AI-PFT-Clin model outperformed the Clin model, showing an area under the receiver operating characteristic curve of 0.755 versus 0.730 (P<.05) for moderate-to-severe AE-COPD and 0.713 versus 0.675 (P<.05) for severe AE-COPD. The AI-PFT-Clin model demonstrated reliable predictive capability across subgroups, including younger patients and those without previous exacerbations. Higher AI-PFT-Clin scores correlated with elevated AE-COPD risk (adjusted hazard ratio for Q4 vs Q1: 4.21, P<.001), with sustained predictive stability over a 10-year follow-up period.

CONCLUSIONS: The AI-PFT-Clin model, by integrating clinical data with spirometry images, offers enhanced predictive accuracy for AE-COPD events compared to a clinical-only approach. This AI-based framework facilitates the early identification of high-risk individuals through the detection of physiological abnormalities not captured by conventional metrics. The model's robust performance and long-term predictive stability suggest its potential utility in proactive COPD management and personalized intervention planning. These findings highlight the promise of incorporating advanced AI techniques into routine COPD management, particularly in populations traditionally seen as lower risk, supporting improved management of COPD through tailored patient care.

PMID:40373296 | DOI:10.2196/69785

ACS Sens. 2025 May 15. doi: 10.1021/acssensors.4c03602. Online ahead of print.

ABSTRACT

The prevalence of sleep disorders in the aging population and the importance of sleep quality for health have emphasized the need for accurate and accessible sleep monitoring solutions. Polysomnography (PSG) remains the clinical gold standard for diagnosing sleep disorders; however, its discomfort and inconvenience limit its accessibility. To address these issues, a wearable device (WD) integrated with stretchable transparent electrodes (STEs) is developed in this study for multisignal sleep monitoring and artificial intelligence (AI)-driven sleep staging. Utilizing conductive and flexible STEs, the WD records multiple biological signals (electroencephalogram [EEG], electrooculogram [EOG], electromyogram [EMG], photoplethysmography, and motion data) with high precision and low noise, comparable to PSG (<4 μVRMS). It achieves a 73.2% accuracy and a macro F1 score of 0.72 in sleep staging using an AI model trained on multisignal inputs. Notably, accuracy marginally improves when using only the EEG, EOG, and EMG channels, which may simplify future device designs. This WD offers a compact, multisignal solution for at-home sleep monitoring, with the potential for use as an evaluation tool for personalized sleep therapies.

PMID:40373282 | DOI:10.1021/acssensors.4c03602

PLoS One. 2025 May 15;20(5):e0322586. doi: 10.1371/journal.pone.0322586. eCollection 2025.

ABSTRACT

Apples are one of the most productive fruits in the world, in addition to their nutritional and health advantages for humans. Even with the continuous development of AI in agriculture in general and apples in particular, automated systems continue to encounter challenges identifying rotten fruit and variations within the same apple category, as well as similarity in type, color, and shape of different fruit varieties. These issues, in addition to apple diseases, substantially impact the economy, productivity, and marketing quality. In this paper, we first provide a novel comprehensive collection named Apple Fruit Varieties Collection (AFVC) with 29,750 images through 85 classes. Second, we distinguish fresh and rotten apples with Apple Fruit Quality Categorization (AFQC), which has 2,320 photos. Third, an Apple Diseases Extensive Collection (ADEC), comprised of 2,976 images with seven classes, was offered. Fourth, following the state of the art, we develop an Optimized Apple Orchard Model (OAOM) with a new loss function named measured focal cross-entropy (MFCE), which assists in improving the proposed model's efficiency. The proposed OAOM gives the highest performance for apple varieties identification with AFVC; accuracy was 93.85%. For the apples rotten recognition with AFQC, accuracy was 98.28%. For the identification of the diseases via ADEC, it was 99.66%. OAOM works with high efficiency and outperforms the baselines. The suggested technique boosts apple system automation with numerous duties and outstanding effectiveness. This research benefits the growth of apple's robotic vision, development policies, automatic sorting systems, and decision-making enhancement.

PMID:40373081 | DOI:10.1371/journal.pone.0322586

PLoS One. 2025 May 15;20(5):e0322197. doi: 10.1371/journal.pone.0322197. eCollection 2025.

ABSTRACT

BACKGROUND: The use of Artificial Intelligence (AI) is exponentially rising in the healthcare sector. This change influences various domains of early identification, diagnosis, and treatment of diseases.

PURPOSE: This study examines the integration of AI in healthcare, focusing on its transformative potential in diagnostics and treatment, and the challenges and methodologies. shaping its future development.

METHODS: The review included 68 academic studies retracted from different databases (WOS, Scopus and Pubmed) from January 2020 and April 2024. After careful review and data analysis, AI methodologies, benefits and challenges, were summarized.

RESULTS: The number of studies showed a steady rise from 2020 to 2023. Most of them were the results of a collaborative work with international universities (92.1%). The majority (66.7%) were published in top-tier (Q1) journals and 40% were cited 2-10 times. The results have shown that AI tools such as deep learning methods and machine learning continue to significantly improve accuracy and timely execution of medical processes. Benefits were discussed from both the organizational and the patient perspective in the categories of diagnosis, treatment, consultation and health monitoring of diseases. However, some challenges may exist, despite these benefits, and are related to data integration, errors related to data processing and decision making, and patient safety.

CONCLUSION: The article examines the present status of AI in medical applications and explores its potential future applications. The findings of this review are useful for healthcare professionals to acquire deeper knowledge on the use of medical AI from design to implementation stage. However, a thorough assessment is essential to gather more insights into whether AI benefits outweigh its risks. Additionally, ethical and privacy issues need careful consideration.

PMID:40372995 | DOI:10.1371/journal.pone.0322197

JMIR Cancer. 2025 May 15;11:e64697. doi: 10.2196/64697.

ABSTRACT

BACKGROUND: Progression-free survival (PFS) is a crucial endpoint in cancer drug research. Clinician-confirmed cancer progression, namely real-world PFS (rwPFS) in unstructured text (ie, clinical notes), serves as a reasonable surrogate for real-world indicators in ascertaining progression endpoints. Response evaluation criteria in solid tumors (RECIST) is traditionally used in clinical trials using serial imaging evaluations but is impractical when working with real-world data. Manual abstraction of clinical progression from unstructured notes remains the gold standard. However, this process is a resource-intensive, time-consuming process. Natural language processing (NLP), a subdomain of machine learning, has shown promise in accelerating the extraction of tumor progression from real-world data in recent years.

OBJECTIVES: We aim to configure a pretrained, general-purpose health care NLP framework to transform free-text clinical notes and radiology reports into structured progression events for studying rwPFS on metastatic breast cancer (mBC) cohorts.

METHODS: This study developed and validated a novel semiautomated workflow to estimate rwPFS in patients with mBC using deidentified electronic health record data from the Nference nSights platform. The developed workflow was validated in a cohort of 316 patients with hormone receptor-positive, human epidermal growth factor receptor-2 (HER-2) 2-negative mBC, who were started on palbociclib and letrozole combination therapy between January 2015 and December 2021. Ground-truth datasets were curated to evaluate the workflow's performance at both the sentence and patient levels. NLP-captured progression or a change in therapy line were considered outcome events, while death, loss to follow-up, and end of the study period were considered censoring events for rwPFS computation. Peak reduction and cumulative decline in Patient Health Questionnaire-8 (PHQ-8) scores were analyzed in the progressed and nonprogressed patient subgroups.

RESULTS: The configured clinical NLP engine achieved a sentence-level progression capture accuracy of 98.2%. At the patient level, initial progression was captured within ±30 days with 88% accuracy. The median rwPFS for the study cohort (N=316) was 20 (95% CI 18-25) months. In a validation subset (n=100), rwPFS determined by manual curation was 25 (95% CI 15-35) months, closely aligning with the computational workflow's 22 (95% CI 15-35) months. A subanalysis revealed rwPFS estimates of 30 (95% CI 24-39) months from radiology reports and 23 (95% CI 19-28) months from clinical notes, highlighting the importance of integrating multiple note sources. External validation also demonstrated high accuracy (92.5% sentence level; 90.2% patient level). Sensitivity analysis revealed stable rwPFS estimates across varying levels of missing source data and event definitions. Peak reduction in PHQ-8 scores during the study period highlighted significant associations between patient-reported outcomes and disease progression.

CONCLUSIONS: This workflow enables rapid and reliable determination of rwPFS in patients with mBC receiving combination therapy. Further validation across more diverse external datasets and other cancer types is needed to ensure broader applicability and generalizability.

PMID:40372953 | DOI:10.2196/64697

IEEE Trans Ultrason Ferroelectr Freq Control. 2025 May 15;PP. doi: 10.1109/TUFFC.2025.3570496. Online ahead of print.

ABSTRACT

Ultrasound localization microscopy (ULM) has revolutionized microvascular imaging by breaking the acoustic diffraction limit. However, different ULM workflows depend heavily on distinct prior knowledge, such as the impulse response and empirical selection of parameters (e.g., the number of microbubbles (MBs) per frame M), or the consistency of training-test dataset in deep learning (DL)-based studies. We hereby propose a general ULM pipeline that reduces priors. Our approach leverages a DL model that simultaneously distills microbubble signals and reduces speckle from every frame without estimating the impulse response and M. Our method features an efficient channel attention vision transformer (ViT) and a progressive learning strategy, enabling it to learn global information through training on progressively increasing patch sizes. Ample synthetic data were generated using the k-Wave toolbox to simulate various MB patterns, thus overcoming the deficiency of labeled data. The ViT output was further processed by a standard radial symmetry method for sub-pixel localization. Our method performed well on model-unseen public datasets: one in silico dataset with ground truth and four in vivo datasets of mouse tumor, rat brain, rat brain bolus, and rat kidney. Our pipeline outperformed conventional ULM, achieving higher positive predictive values (precision in DL, 0.88-0.41 vs. 0.83-0.16) and improved accuracy (root-mean-square errors: 0.25-0.14 λ vs. 0.31-0.13 λ) across a range of signal-to-noise ratios from 60 dB to 10 dB. Our model could detect more vessels in diverse in vivo datasets while achieving comparable resolutions to the standard method. The proposed ViT-based model, seamlessly integrated with state-of-the-art downstream ULM steps, improved the overall ULM performance with no priors.

PMID:40372868 | DOI:10.1109/TUFFC.2025.3570496

IEEE Trans Neural Netw Learn Syst. 2025 May 15;PP. doi: 10.1109/TNNLS.2025.3566021. Online ahead of print.

ABSTRACT

Speech enhancement (SE) is critical for improving speech intelligibility and quality in various audio devices. In recent years, deep learning-based methods have significantly improved SE performance, but they often come with a high computational cost, which is prohibitive for a large number of edge devices, such as headsets and hearing aids. This work proposes an ultralow-power SE system based on the brain-inspired spiking neural network (SNN) called Spiking-FullSubNet. Spiking-FullSubNet follows a full-band and subband fusioned approach to effectively capture both global and local spectral information. To enhance the efficiency of computationally expensive subband modeling, we introduce a frequency partitioning method inspired by the sensitivity profile of the human peripheral auditory system. Furthermore, we introduce a novel spiking neuron model that can dynamically control the input information integration and forgetting, enhancing the multiscale temporal processing capability of SNN, which is critical for speech denoising. Experiments conducted on the recent Intel Neuromorphic Deep Noise Suppression (N-DNS) Challenge dataset show that the Spiking-FullSubNet surpasses state-of-the-art (SOTA) methods by large margins in terms of both speech quality and energy efficiency metrics. Notably, our system won the championship of the Intel N-DNS Challenge (algorithmic track), opening up a myriad of opportunities for ultralow-power SE at the edge. Our source code and model checkpoints are publicly available at github.com/haoxiangsnr/spiking-fullsubnet.

PMID:40372867 | DOI:10.1109/TNNLS.2025.3566021

IEEE Trans Med Imaging. 2025 May 15;PP. doi: 10.1109/TMI.2025.3570342. Online ahead of print.

ABSTRACT

Positron Emission Tomography (PET) is an important clinical imaging tool but inevitably introduces radiation exposure to patients and healthcare providers. Reducing the tracer injection dose and eliminating the CT acquisition for attenuation correction can reduce the overall radiation dose, but often results in PET with high noise and bias. Thus, it is desirable to develop 3D methods to translate the non-attenuation-corrected low-dose PET (NAC-LDPET) into attenuation-corrected standard-dose PET (AC-SDPET). Recently, diffusion models have emerged as a new state-of-the-art deep learning method for image-to-image translation, better than traditional CNN-based methods. However, due to the high computation cost and memory burden, it is largely limited to 2D applications. To address these challenges, we developed a novel 2.5D Multi-view Averaging Diffusion Model (MADM) for 3D image-to-image translation with application on NAC-LDPET to AC-SDPET translation. Specifically, MADM employs separate diffusion models for axial, coronal, and sagittal views, whose outputs are averaged in each sampling step to ensure the 3D generation quality from multiple views. To accelerate the 3D sampling process, we also proposed a strategy to use the CNN-based 3D generation as a prior for the diffusion model. Our experimental results on human patient studies suggested that MADM can generate high-quality 3D translation images, outperforming previous CNN-based and Diffusion-based baseline methods. The code is available at https://github.com/tianqic/MADM.

PMID:40372846 | DOI:10.1109/TMI.2025.3570342

Gigascience. 2025 Jan 6;14:giaf037. doi: 10.1093/gigascience/giaf037.

ABSTRACT

BACKGROUND: Bacteriophages are among the most abundant organisms on Earth, significantly impacting ecosystems and human society. The identification of viral sequences, especially novel ones, from mixed metagenomes is a critical first step in analyzing the viral components of host samples. This plays a key role in many downstream tasks. However, this is a challenging task due to their rapid evolution rate. The identification process typically involves two steps: distinguishing viral sequences from the host and identifying if they come from novel viral genomes. Traditional metagenomic techniques that rely on sequence similarity with known entities often fall short, especially when dealing with short or novel genomes. Meanwhile, deep learning has demonstrated its efficacy across various domains, including the bioinformatics field.

RESULTS: We have developed HVSeeker-a host/virus seeker method-based on deep learning to distinguish between bacterial and phage sequences. HVSeeker consists of two separate models: one analyzing DNA sequences and the other focusing on proteins. In addition to the robust architecture of HVSeeker, three distinct preprocessing methods were introduced to enhance the learning process: padding, contigs assembly, and sliding window. This method has shown promising results on sequences with various lengths, ranging from 200 to 1,500 base pairs. Tested on both NCBI and IMGVR databases, HVSeeker outperformed several methods from the literature such as Seeker, Rnn-VirSeeker, DeepVirFinder, and PPR-Meta. Moreover, when compared with other methods on benchmark datasets, HVSeeker has shown better performance, establishing its effectiveness in identifying unknown phage genomes.

CONCLUSIONS: These results demonstrate the exceptional structure of HVSeeker, which encompasses both the preprocessing methods and the model design. The advancements provided by HVSeeker are significant for identifying viral genomes and developing new therapeutic approaches, such as phage therapy. Therefore, HVSeeker serves as an essential tool in prokaryotic and phage taxonomy, offering a crucial first step toward analyzing the host-viral component of samples by identifying the host and viral sequences in mixed metagenomes.

PMID:40372723 | DOI:10.1093/gigascience/giaf037

Int J Comput Assist Radiol Surg. 2025 May 15. doi: 10.1007/s11548-025-03378-1. Online ahead of print.

ABSTRACT

PURPOSE: Deep learning has been widely used in research on sparse-view computed tomography (CT) image reconstruction. While sufficient training data can lead to high accuracy, collecting medical images is often challenging due to legal or ethical concerns, making it necessary to develop methods that perform well with limited data. To address this issue, we explored the use of nonmedical images for pre-training. Therefore, in this study, we investigated whether fractal images could improve the quality of sparse-view CT images, even with a reduced number of medical images.

METHODS: Fractal images generated by an iterated function system (IFS) were used for nonmedical images, and medical images were obtained from the CHAOS dataset. Sinograms were then generated using 36 projections in sparse-view and the images were reconstructed by filtered back-projection (FBP). FBPConvNet and WNet (first module: learning fractal images, second module: testing medical images, and third module: learning output) were used as networks. The effectiveness of pre-training was then investigated for each network. The quality of the reconstructed images was evaluated using two indices: structural similarity (SSIM) and peak signal-to-noise ratio (PSNR).

RESULTS: The network parameters pre-trained with fractal images showed reduced artifacts compared to the network trained exclusively with medical images, resulting in improved SSIM. WNet outperformed FBPConvNet in terms of PSNR. Pre-training WNet with fractal images produced the best image quality, and the number of medical images required for main-training was reduced from 5000 to 1000 (80% reduction).

CONCLUSION: Using fractal images for network training can reduce the number of medical images required for artifact reduction in sparse-view CT. Therefore, fractal images can improve accuracy even with a limited amount of training data in deep learning.

PMID:40372595 | DOI:10.1007/s11548-025-03378-1

Bioinformatics. 2025 May 15:btaf294. doi: 10.1093/bioinformatics/btaf294. Online ahead of print.

ABSTRACT

MOTIVATION: Allostery, the process by which binding at one site perturbs a distant site, is being rendered as a key focus in the field of drug development with its substantial impact on protein function. The identification of allosteric pockets (sites) is a challenging task and several techniques have been developed, including Machine Learning (ML) to predict allosteric pockets that utilize both static and pocket features.

RESULTS: Our work, DeepAllo, is the first study that combines fine-tuned protein language model (pLM) with FPocket features and shows an increase in prediction performance of allosteric sites over previous studies. The pLM model was fine-tuned on Allosteric Dataset (ASD) in Multitask Learning (MTL) setting and was further used as a feature extractor to train XGBoost and AutoML models. The best model predicts allosteric pockets with 89.66% F1 score and 90.5% of allosteric pockets in the top 3 positions, outperforming previous results. A case study has been performed on proteins with known allosteric pockets, which shows the proof of our approach. Moreover, an effort was made to explain the pLM by visualizing its attention mechanism among allosteric and non-allosteric residues.

AVAILABILITY: The source code is available on GitHub (https://github.com/MoaazK/deepallo) and archived on Zenodo (DOI: 10.5281/zenodo.15255379). The trained model is hosted on Hugging Face (DOI: 10.57967/hf/5198). The dataset used for training and evaluation is archived on Zenodo (DOI: 10.5281/zenodo.15255437).

SUPPLEMENTARY INFORMATION: Supplementary data, including the full list of proteins used in the study with their PDB IDs, t-SNE analysis of pocket features, confusion matrix breakdown, and interpretation of borderline classification cases are available as supplementary material along this article.

PMID:40372465 | DOI:10.1093/bioinformatics/btaf294

Osteoporos Int. 2025 May 15. doi: 10.1007/s00198-025-07515-z. Online ahead of print.

ABSTRACT

Video-estimated peak jump power (vJP) using deep learning showed strong agreement with ground truth jump power (gJP). vJP was associated with sarcopenia, age, and muscle parameters in adults, with providing a proof-of-concept that markerless monitoring of peak jump power could be feasible in daily life space.

OBJECTIVES: Low peak countermovement jump power measured by ground force plate (GFP) is associated with sarcopenia, impaired physical function, and elevated risk of fracture in older adults. GFP is available at research setting yet, limiting its clinical applicability. Video-based estimation of peak jump power could enhance clinical applicability of jump power measurement over research setting.

METHODS: Data were collected prospectively in osteoporosis clinic of Severance Hospital, Korea, between March and August 2022. Individuals performed three jump attempts on GFP (ground truth, gJP) under video recording, along with measurement of handgrip strength (HGS), 5-time chair rise (CRT) test, and appendicular lean mass (ALM). Open source deep learning pose estimation and machine learning algorithms were used to estimate video-estimated peak jump power (vJP) in 80% train set. Sarcopenia was defined by Korean Working Group for Sarcopenia 2023 definition.

RESULTS: A total of 658 jump motion data from 220 patients (mean age 62 years; 77% women; sarcopenia 19%) were analyzed. In test set (20% hold-out set), average difference between predicted and actual jump power was 0.27 W/kg (95% limit of agreement - 5.01 to + 5.54 W/kg; correlation coefficient 0.93). vJP detected gJP-defined low jump power with 81.8% sensitivity and 91.3% specificity. vJP showed a steep decline across age like gJP, with modest to strong correlation with HGS and CRT. Eight landmarks (both shoulders, hip, knee joints, and ears) were the most contributing features to vJP estimation. vJP was associated with the presence of sarcopenia (unadjusted and adjusted, - 3.95 and - 2.30 W/kg), HGS (- 3.69 and - 1.96 W/kg per 1 SD decrement), and CRT performance (- 2.79 and - 1.87 W/kg per 1 SD decrement in log-CRT) similar to that of gJP.

CONCLUSION: vJP was associated with sarcopenia, age, and muscle parameters in adults, with good agreement with ground truth jump power.

PMID:40372459 | DOI:10.1007/s00198-025-07515-z

Eur Spine J. 2025 May 15. doi: 10.1007/s00586-025-08900-2. Online ahead of print.

ABSTRACT

PURPOSE: To assess the agreement between lumbar disc degeneration (DD) grading by the convolutional neural network model SpineNet and radiologist's visual grading.

METHODS: In a 14-year follow-up MRI study involving 19 male volunteers, lumbar DD was assessed by SpineNet and two radiologists using the Pfirrmann classification at baseline (age 37) and after 14 years (age 51). Pfirrmann summary scores (PSS) were calculated by summing individual disc grades. The agreement between the first radiologist and SpineNet was analyzed, with the second radiologist's grading used for inter-observer agreement.

RESULTS: Significant differences were observed in the Pfirrmann grades and PSS assigned by the radiologist and SpineNet at both time points. SpineNet assigned Pfirrmann grade 1 to several discs and grade 5 to more discs compared to the radiologists. The concordance correlation coefficients (CCC) of PSS between the radiologist and SpineNet were 0.54 (95% CI: 0.28 to 0.79) at baseline and 0.54 (0.27 to 0.80) at follow-up. The average kappa (κ) values of 0.74 (0.68 to 0.81) at baseline and 0.68 (0.58 to 0.77) at follow-up. CCC of PSS between the radiologists was 0.83 (0.69 to 0.97) at baseline and 0.78 (0.61 to 0.95) at follow-up, with κ values ranging from 0.73 to 0.96.

CONCLUSION: We found fair to substantial agreement in DD grading between SpineNet and the radiologist, albeit with notable discrepancies. These findings indicate that AI-based systems like SpineNet hold promise as complementary tools in radiological evaluation, including in longitudinal studies, but emphasize the need for ongoing refinement of AI algorithms.

PMID:40372457 | DOI:10.1007/s00586-025-08900-2

ACS Sens. 2025 May 15. doi: 10.1021/acssensors.5c00630. Online ahead of print.

ABSTRACT

Mimicking the olfactory system of humans, the use of electronic noses (E-noses) for the detection of odors in nature has become a hot research topic. This study presents a novel E-nose based on deep learning architecture called Scentformer, which addresses the limitations of the current E-nose like a narrow detection range and limited generalizability across different scenarios. Armed with a self-adaptive data down-sampling method, the E-nose is capable of detecting 55 different natural odors with the classification accuracy of 99.94%, and the model embedded in the E-nose is analyzed using Shapley Additive exPlanations analysis, providing a quantitative interpretation of the E-nose performance. Furthermore, leveraging Scentformer's transfer learning ability, the E-nose efficiently adapts to new odors and gases. Rather than retraining all layers of the model on the new odor data set, only the fully connected layers need to be trained for the pretrained model. Using only 1‰ data of the retrained model, the pretrained model-based E-nose can also achieve classification accuracies of 99.14% across various odor and gas concentrations. This provides a robust approach to the detection of diverse direct current signals in real-world applications.

PMID:40372408 | DOI:10.1021/acssensors.5c00630

Hum Brain Mapp. 2025 Jun 1;46(8):e70229. doi: 10.1002/hbm.70229.

ABSTRACT

Deformable brain templates are an important tool in many neuroimaging analyses. Conditional templates (e.g., age-specific templates) have advantages over single population templates by enabling improved registration accuracy and capturing common processes in brain development and degeneration. Conventional methods require large, evenly spread cohorts to develop conditional templates, limiting their ability to create templates that could reflect richer combinations of clinical and demographic variables. More recent deep-learning methods, which can infer relationships in very high-dimensional spaces, open up the possibility of producing conditional templates that are jointly optimised for these richer sets of conditioning parameters. We have built on recent deep-learning template generation approaches using a diffeomorphic (topology-preserving) framework to create a purely geometric method of conditional template construction that learns diffeomorphisms between: (i) a global or group template and conditional templates, and (ii) conditional templates and individual brain scans. We evaluated our method, as well as other recent deep-learning approaches, on a data set of cognitively normal (CN) participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI), using age as the conditioning parameter of interest. We assessed the effectiveness of these networks at capturing age-dependent anatomical differences. Our results demonstrate that while the assessed deep-learning methods have a number of strengths, they require further refinement to capture morphological changes in ageing brains with an acceptable degree of accuracy. The volumetric output of our method, and other recent deep-learning approaches, across four brain structures (grey matter, white matter, the lateral ventricles and the hippocampus), was measured and showed that although each of the methods captured some changes well, each method was unable to accurately track changes in all of the volumes. However, as our method is purely geometric, it was able to produce T1-weighted conditional templates with high spatial fidelity and with consistent topology as age varies, making these conditional templates advantageous for spatial registrations. The use of diffeomorphisms in these deep-learning methods represents an important strength of these approaches, as they can produce conditional templates that can be explicitly linked, geometrically, across age as well as to fixed, unconditional templates or brain atlases. The use of deep learning in conditional template generation provides a framework for creating templates for more complex sets of conditioning parameters, such as pathologies and demographic variables, in order to facilitate a broader application of conditional brain templates in neuroimaging studies. This can aid researchers and clinicians in their understanding of how brain structure changes over time and under various interventions, with the ultimate goal of improving the calibration of treatments and interventions in personalised medicine. The code to implement our conditional brain template network is available at: github.com/lwhitbread/deep-diff.

PMID:40372124 | DOI:10.1002/hbm.70229