Magn Reson Med. 2025 Feb 27. doi: 10.1002/mrm.30478. Online ahead of print.

ABSTRACT

PURPOSE: The clinical value of quantitative MRI hinges on its measurement repeatability. Deep learning methods to reconstruct undersampled quantitative MRI can accelerate reconstruction but do not aim to promote quantitative repeatability. This study proposes a repeatability-encouraging self-supervised learning (SSL) reconstruction method for quantitative MRI.

METHODS: The proposed SSL reconstruction network minimized cross-data-consistency between two equally sized, mutually exclusive temporal subsets of k-t-space data, encouraging repeatability by enabling each subset's reconstruction to predict the other's k-t-space data. The method was evaluated on cardiac MR Multitasking T1 mapping data and compared with supervised learning methods trained on full 60-s inputs (Sup60) and on split 30-s inputs (Sup30/30). Reconstruction quality was evaluated on full 60-s inputs, comparing results to iterative wavelet-regularized references using Bland-Altman limits of agreement (LOAs). Repeatability was evaluated by splitting the 60-s data into two 30-s inputs, evaluating T1 differences between reconstructions from the two halves of the scan.

RESULTS: On 60-s inputs, the proposed method produced comparable-quality images and T1 maps to the Sup60 method, with T1 values in general agreement with the wavelet reference (LOA Sup60 = ±75 ms, SSL = ±81 ms), whereas the Sup30/30 method generated blurrier results and showed poor T1 agreement (LOA Sup30/30 = ±132 ms). On back-to-back 30-s inputs, the proposed method had the best T1 repeatability (coefficient of variation SSL = 6.3%, Sup60 = 12.0%, Sup30/30 = 6.9%). Of the three deep learning methods, only the SSL method produced sharp and repeatable images.

CONCLUSION: Without the need for labeled training data, the proposed SSL method demonstrated superior repeatability compared with supervised learning without sacrificing sharpness, and reduced reconstruction time versus iterative methods.

PMID:40014485 | DOI:10.1002/mrm.30478

Transl Vis Sci Technol. 2025 Feb 3;14(2):26. doi: 10.1167/tvst.14.2.26.

ABSTRACT

PURPOSE: This study aims to develop an automated pipeline to detect retinal detachment from B-scan ocular ultrasonography (USG) images by using deep learning-based segmentation.

METHODS: A computational pipeline consisting of an encoder-decoder segmentation network and a machine learning classifier was developed, trained, and validated using 279 B-scan ocular USG images from 204 patients, including 66 retinal detachment (RD) images, 36 posterior vitreous detachment images, and 177 healthy control images. Performance metrics, including the precision, recall, and F-scores, were calculated for both segmentation and RD detection.

RESULTS: The overall pipeline achieved 96.3% F-score for RD detection, outperforming end-to-end deep learning classification models (ResNet-50 and MobileNetV3) with 94.3% and 95.0% F-scores. This improvement was also validated on an independent test set, where the proposed pipeline led to 96.5% F-score, but the classification models yielded only 62.1% and 84.9% F-scores, respectively. Besides, the segmentation model of this pipeline led to high performances across multiple ocular structures, with 84.7%, 78.3%, and 88.2% F-scores for retina/choroid, sclera, and optic nerve sheath segmentation, respectively. The segmentation model outperforms the standard UNet, particularly in challenging RD cases, where it effectively segmented detached retina regions.

CONCLUSIONS: The proposed automated segmentation and classification method improves RD detection in B-scan ocular USG images compared to end-to-end classification models, offering potential clinical benefits in resource-limited settings.

TRANSLATIONAL RELEVANCE: We have developed a novel deep/machine learning based pipeline that has the potential to significantly improve diagnostic accuracy and accessibility for ocular USG.

PMID:40014336 | DOI:10.1167/tvst.14.2.26

Eur J Nucl Med Mol Imaging. 2025 Feb 27. doi: 10.1007/s00259-025-07149-7. Online ahead of print.

ABSTRACT

PURPOSE: PET/CT imaging data contains a wealth of quantitative information that can provide valuable contributions to characterising tumours. A growing body of work focuses on the use of deep-learning (DL) techniques for denoising PET data. These models are clinically evaluated prior to use, however, quantitative image assessment provides potential for further evaluation. This work uses radiomic features to compare two manufacturer deep-learning (DL) image enhancement algorithms, one of which has been commercialised, against 'gold-standard' image reconstruction techniques in phantom data and a sarcoma patient data set (N=20).

METHODS: All studies in the retrospective sarcoma clinical [ 18 F]FDG dataset were acquired on either a GE Discovery 690 or 710 PET/CT scanner with volumes segmented by an experienced nuclear medicine radiologist. The modular heterogeneous imaging phantom used in this work was filled with [ 18 F]FDG, and five repeat acquisitions of the phantom were acquired on a GE Discovery 710 PET/CT scanner. The DL-enhanced images were compared to 'gold-standard' images the algorithms were trained to emulate and input images. The difference between image sets was tested for significance in 93 international biomarker standardisation initiative (IBSI) standardised radiomic features.

RESULTS: Comparing DL-enhanced images to the 'gold-standard', 4.0% and 9.7% radiomic features measured significantly different (pcritical < 0.0005) in the phantom and patient data respectively (averaged over the two DL algorithms). Larger differences were observed comparing DL-enhanced images to algorithm input images with 29.8% and 43.0% of radiomic features measuring significantly different in the phantom and patient data respectively (averaged over the two DL algorithms).

CONCLUSION: DL-enhanced images were found to be similar to images generated using the 'gold-standard' target image reconstruction method with more than 80% of radiomic features not significantly different in all comparisons across unseen phantom and sarcoma patient data. This result offers insight into the performance of the DL algorithms, and demonstrate potential applications for DL algorithms in harmonisation for radiomics and for radiomic features in quantitative evaluation of DL algorithms.

PMID:40014074 | DOI:10.1007/s00259-025-07149-7

J Appl Clin Med Phys. 2025 Feb 27:e70034. doi: 10.1002/acm2.70034. Online ahead of print.

ABSTRACT

Artificial intelligence (AI) has the potential to revolutionize brachytherapy's clinical workflow. This review comprehensively examines the application of AI, focusing on machine learning and deep learning, in various aspects of brachytherapy. We analyze AI's role in making brachytherapy treatments more personalized, efficient, and effective. The applications are systematically categorized into seven categories: imaging, preplanning, treatment planning, applicator reconstruction, quality assurance, outcome prediction, and real-time monitoring. Each major category is further subdivided based on cancer type or specific tasks, with detailed summaries of models, data sizes, and results presented in corresponding tables. Additionally, we discuss the limitations, challenges, and ethical concerns of current AI applications, along with perspectives on future directions. This review offers insights into the current advancements, challenges, and the impact of AI on treatment paradigms, encouraging further research to expand its clinical utility.

PMID:40014044 | DOI:10.1002/acm2.70034

J Neurophysiol. 2025 Feb 27. doi: 10.1152/jn.00622.2024. Online ahead of print.

ABSTRACT

Dendritic spines, small protrusions on neuronal dendrites, play a crucial role in brain function by changing shape and size in response to neural activity. So far, in depth analysis of dendritic spines in human brain tissue is lacking. This study presents a comprehensive analysis of human dendritic spine morphology and density using a unique dataset from human brain tissue from 27 patients (8 females, 19 males, aged 18-71) undergoing tumor or epilepsy surgery at three neurosurgery sites. We used acute slices and organotypic brain slice cultures to examine dendritic spines, classifying them into the three main morphological subtypes: Mushroom, Thin, and Stubby, via 3D reconstruction using ZEISS arivis Pro software. A deep learning model, trained on 39 diverse datasets, automated spine segmentation and 3D reconstruction, achieving a 74% F1-score and reducing processing time by over 50%. We show significant differences in spine density by sex, dendrite type, and tissue condition. Females had higher spine densities than males, and apical dendrites were denser in spines than basal ones. Acute tissue showed higher spine densities compared to cultured human brain tissue. With time in culture, Mushroom spines decreased, while Stubby and Thin spine percentages increased, particularly from 7-9 to 14 days in vitro, reflecting potential synaptic plasticity changes. Our study underscores the importance of using human brain tissue to understand unique synaptic properties and shows that integrating deep learning with traditional methods enables efficient large-scale analysis, revealing key insights into sex- and tissue-specific dendritic spine dynamics relevant to neurological diseases.

PMID:40013734 | DOI:10.1152/jn.00622.2024

J Chem Inf Model. 2025 Feb 27. doi: 10.1021/acs.jcim.4c02051. Online ahead of print.

ABSTRACT

Cytochrome P450 enzymes (CYP450s) play crucial roles in metabolizing many drugs, and thus, local chemical structure can profoundly influence drug efficacy and toxicity. Therefore, the accurate prediction of CYP450-mediated reaction sites can increase the efficiency of drug discovery and development. Here, we present GLMCyp, a deep learning-based approach, for predicting CYP450 reaction sites on small molecules. By integrating two-dimensional (2D) molecular graph features, three-dimensional (3D) features from Uni-Mol, and relevant CYP450 protein features generated by ESM-2, GLMCyp could accurately predict bonds of metabolism (BoMs) targeted by a panel of nine human CYP450s. Incorporating protein features allowed GLMCyp application in broader CYP450 metabolism prediction tasks. Additionally, substrate molecular feature processing enhanced the accuracy and interpretability of the predictions. The model was trained on the EBoMD data set and reached an area under the receiver operating characteristic curve (ROC-AUC) of 0.926. GLMCyp also showed a relatively strong capacity for feature extraction and generalizability in validation with external data sets. The GLMCyp model and data sets are available for public use (https://github.com/lvimmind/GLMCyp-Predictor) to facilitate drug metabolism screening.

PMID:40013456 | DOI:10.1021/acs.jcim.4c02051

Front Oncol. 2025 Feb 12;15:1564325. doi: 10.3389/fonc.2025.1564325. eCollection 2025.

ABSTRACT

[This corrects the article DOI: 10.3389/fonc.2024.1458374.].

PMID:40012552 | PMC:PMC11862821 | DOI:10.3389/fonc.2025.1564325

J Thorac Imaging. 2025 Feb 27. doi: 10.1097/RTI.0000000000000827. Online ahead of print.

ABSTRACT

PURPOSE: To assess the performance of a newly introduced deep learning-based reconstruction algorithm, namely the artificial intelligence iterative reconstruction (AIIR), in reducing the dose of pediatric chest CT by using the image data of below 3-year-old patients with congenital heart disease (CHD).

MATERIALS AND METHODS: The lung image available from routine-dose cardiac CT angiography (CTA) on below 3 years patients with CHD was employed as a reference for evaluating the paired low-dose chest CT. A total of 191 subjects were prospectively enrolled, where the dose for chest CT was reduced to ~0.1 mSv while the cardiac CTA protocol was kept unchanged. The low-dose chest CT images, obtained with the AIIR and the hybrid iterative reconstruction (HIR), were compared in image quality, ie, overall image quality and lung structure depiction, and in diagnostic performance, ie, severity assessment of pneumonia and airway stenosis.

RESULTS: Compared with the reference, lung image quality was not found significantly different on low-dose AIIR images (all P>0.05) but obviously inferior with the HIR (all P<0.05). Compared with the HIR, low-dose AIIR images also achieved a closer pneumonia severity index (AIIR 4.32±3.82 vs. Ref 4.37±3.84, P>0.05; HIR 5.12±4.06 vs. Ref 4.37±3.84, P<0.05) and airway stenosis grading (consistently graded: AIIR 88.5% vs. HIR 56.5% ) to the reference.

CONCLUSIONS: AIIR has the potential for large dose reduction in chest CT of patients below 3 years of age while preserving image quality and achieving diagnostic results nearly equivalent to routine dose scans.

PMID:40013381 | DOI:10.1097/RTI.0000000000000827

J Evid Based Med. 2025 Mar;18(1):e70005. doi: 10.1111/jebm.70005.

ABSTRACT

OBJECTIVE: Leukemia is a type of blood cancer that begins in the bone marrow and results in high numbers of abnormal white blood cells. Automated detection and classification of leukemia and its subtypes using artificial intelligence (AI) and machine learning (ML) algorithms plays a significant role in the early diagnosis and treatment of this fatal disease. This study aimed to review and synthesize research findings on AI-based approaches in leukemia detection and classification from peripheral blood smear images.

METHODS: A systematic literature search was conducted across four e-databases (Web of Science, PubMed, Scopus, and IEEE Xplore) from January 2015 to March 2023 by searching the keywords "Leukemia," "Machine Learning," and "Blood Smear Image," as well as their synonyms. All original journal articles and conference papers that used ML algorithms in detecting and classifying leukemia were included. The study quality was assessed using the Qiao Quality Assessment tool.

RESULTS: From 1325 articles identified through a systematic search, 190 studies were eligible for this review. The mean validation accuracy (ACC) of the ML methods applied in the reviewed studies was 95.38%. Among different ML methods, modern techniques were mostly considered to detect and classify leukemia (60.53% of studies). Supervised learning was the dominant ML paradigm (79% of studies). Studies utilized common ML methodologies for leukemia detection and classification, including preprocessing, feature extraction, feature selection, and classification. Deep learning (DL) techniques, especially convolutional neural networks, were the most widely used modern algorithms in the mentioned methodologies. Most studies relied on internal validation (87%). Moreover, K-fold cross-validation and train/test split were the commonly employed validation strategies.

CONCLUSION: AI-based algorithms are widely used in detecting and classifying leukemia with remarkable performance. Future studies should prioritize rigorous external validation to evaluate generalizability.

PMID:40013326 | DOI:10.1111/jebm.70005

IEEE Int Conf Connect Health Appl Syst Eng Technol. 2023 Jun;2023:68-79. doi: 10.1145/3580252.3586974. Epub 2024 Jan 22.

ABSTRACT

Social presence has been known to impact eating behavior among people with obesity; however, the dual study of eating behavior and social presence in real-world settings is challenging due to the inability to reliably confirm the co-occurrence of these important factors. High-resolution video cameras can detect timing while providing visual confirmation of behavior; however, their potential to capture all-day behavior is limited by short battery lifetime and lack of autonomy in detection. Low-resolution infrared (IR) sensors have shown promise in automating human behavior detection; however, it is unknown if IR sensors contribute to behavior detection when combined with RGB cameras. To address these challenges, we designed and deployed a low-power, and low-resolution RGB video camera, in conjunction with a low-resolution IR sensor, to test a learned model's ability to detect eating and social presence. We evaluated our system in the wild with 10 participants with obesity; our models displayed slight improvement when detecting eating (5%) and significant improvement when detecting social presence (44%) compared with using a video-only approach. We analyzed device failure scenarios and their implications for future wearable camera design and machine learning pipelines. Lastly, we provide guidance for future studies using low-cost RGB and IR sensors to validate human behavior with context.

PMID:40013103 | PMC:PMC11864367 | DOI:10.1145/3580252.3586974

Open Mind (Camb). 2025 Feb 16;9:305-324. doi: 10.1162/opmi_a_00189. eCollection 2025.

ABSTRACT

Humans make rich inferences about the geometry of the visual world. While deep neural networks (DNNs) achieve human-level performance on some psychophysical tasks (e.g., rapid classification of object or scene categories), they often fail in tasks requiring inferences about the underlying shape of objects or scenes. Here, we ask whether and how this gap in 3D shape representation between DNNs and humans can be closed. First, we define the problem space: after generating a stimulus set to evaluate 3D shape inferences using a match-to-sample task, we confirm that standard DNNs are unable to reach human performance. Next, we construct a set of candidate 3D-aware DNNs including 3D neural field (Light Field Network), autoencoder, and convolutional architectures. We investigate the role of the learning objective and dataset by training single-view (the model only sees one viewpoint of an object per training trial) and multi-view (the model is trained to associate multiple viewpoints of each object per training trial) versions of each architecture. When the same object categories appear in the model training and match-to-sample test sets, multi-view DNNs approach human-level performance for 3D shape matching, highlighting the importance of a learning objective that enforces a common representation across viewpoints of the same object. Furthermore, the 3D Light Field Network was the model most similar to humans across all tests, suggesting that building in 3D inductive biases increases human-model alignment. Finally, we explore the generalization performance of multi-view DNNs to out-of-distribution object categories not seen during training. Overall, our work shows that multi-view learning objectives for DNNs are necessary but not sufficient to make similar 3D shape inferences as humans and reveals limitations in capturing human-like shape inferences that may be inherent to DNN modeling approaches. We provide a methodology for understanding human 3D shape perception within a deep learning framework and highlight out-of-domain generalization as the next challenge for learning human-like 3D representations with DNNs.

PMID:40013087 | PMC:PMC11864798 | DOI:10.1162/opmi_a_00189

Digit Health. 2025 Feb 16;11:20552076251320726. doi: 10.1177/20552076251320726. eCollection 2025 Jan-Dec.

ABSTRACT

OBJECTIVE: To develop an enhanced deep learning model, MRpoxNet, based on a modified ResNet50 architecture for the early detection of monkeypox from digital skin lesion images, ensuring high diagnostic accuracy and clinical reliability.

METHODS: The study utilized the Kaggle MSID dataset, initially comprising 1156 images, augmented to 6116 images across three classes: monkeypox, non-monkeypox, and normal skin. MRpoxNet was developed by extending ResNet50 from 177 to 182 layers, incorporating additional convolutional, ReLU, dropout, and batch normalization layers. Performance was evaluated using metrics such as accuracy, precision, recall, F1 score, sensitivity, and specificity. Comparative analyses were conducted against established models like ResNet50, AlexNet, VGG16, and GoogleNet.

RESULTS: MRpoxNet achieved a diagnostic accuracy of 98.1%, outperforming baseline models in all key metrics. The enhanced architecture demonstrated superior robustness in distinguishing monkeypox lesions from other skin conditions, highlighting its potential for reliable clinical application.

CONCLUSION: MRpoxNet provides a robust and efficient solution for early monkeypox detection. Its superior performance suggests readiness for integration into diagnostic workflows, with future enhancements aimed at dataset expansion and multimodal adaptability to diverse clinical scenarios.

PMID:40013075 | PMC:PMC11863262 | DOI:10.1177/20552076251320726

Biophys Rev (Melville). 2025 Feb 21;6(1):011307. doi: 10.1063/5.0249920. eCollection 2025 Mar.

ABSTRACT

Accurately predicting mutation-caused binding free energy changes (ΔΔGs) on protein interactions is crucial for understanding how genetic variations affect interactions between proteins and other biomolecules, such as proteins, DNA/RNA, and ligands, which are vital for regulating numerous biological processes. Developing computational approaches with high accuracy and efficiency is critical for elucidating the mechanisms underlying various diseases, identifying potential biomarkers for early diagnosis, and developing targeted therapies. This review provides a comprehensive overview of recent advancements in predicting the impact of mutations on protein interactions across different interaction types, which are central to understanding biological processes and disease mechanisms, including cancer. We summarize recent progress in predictive approaches, including physicochemical-based, machine learning, and deep learning methods, evaluating the strengths and limitations of each. Additionally, we discuss the challenges related to the limitations of mutational data, including biases, data quality, and dataset size, and explore the difficulties in developing accurate prediction tools for mutation-induced effects on protein interactions. Finally, we discuss future directions for advancing these computational tools, highlighting the capabilities of advancing technologies, such as artificial intelligence to drive significant improvements in mutational effects prediction.

PMID:40013003 | PMC:PMC11857871 | DOI:10.1063/5.0249920

Front Med (Lausanne). 2025 Feb 12;12:1507258. doi: 10.3389/fmed.2025.1507258. eCollection 2025.

ABSTRACT

INTRODUCTION: Pulmonary granulomatous nodules (PGN) often exhibit similar CT morphological features to solid lung adenocarcinomas (SLA), making preoperative differentiation challenging. This study aims to address this diagnostic challenge by developing a novel deep learning model.

METHODS: This study proposes MAEMC-NET, a model integrating generative (Masked AutoEncoder) and contrastive (Momentum Contrast) self-supervised learning to learn CT image representations of intra- and inter-solitary nodules. A generative self-supervised task of reconstructing masked axial CT patches containing lesions was designed to learn intra- and inter-slice image representations. Contrastive momentum is used to link the encoder in axial-CT-patch path with the momentum encoder in coronal-CT-patch path. A total of 494 patients from two centers were included.

RESULTS: MAEMC-NET achieved an area under curve (95% Confidence Interval) of 0.962 (0.934-0.973). These results not only significantly surpass the joint diagnosis by two experienced chest radiologists (77.3% accuracy) but also outperform the current state-of-the-art methods. The model performs best on medical images with a 50% mask ratio, showing a 1.4% increase in accuracy compared to the optimal 75% mask ratio on natural images.

DISCUSSION: The proposed MAEMC-NET effectively distinguishes between benign and malignant solitary pulmonary nodules and holds significant potential to assist radiologists in improving the diagnostic accuracy of PGN and SLA.

PMID:40012977 | PMC:PMC11861088 | DOI:10.3389/fmed.2025.1507258

IEEE Robot Autom Lett. 2024 May;9(5):3988-3994. doi: 10.1109/lra.2024.3374189. Epub 2024 Mar 6.

ABSTRACT

Deep learning offers a potentially powerful alternative to hand-tuned control of active lower-limb prostheses, being capable of generating continuous joint-level assistance end-to-end. This eliminates the need for conventional task classification, state machines and mid-level control equations by collapsing the entire control problem into a deep neural network. In this letter, sensor data and conventional commanded torque from an open-source powered knee-ankle prosthesis (OSL) were collected across five locomotion modes: level ground, ramp incline/decline and stair ascent/descent. Reference commanded torques were generated using an expert-tuned finite state machine-based impedance controller for each mode and transfemoral amputee participant (N = 12). Stance phases of the output were then estimated using a temporal convolutional network (TCN), which produced mode- and user-independent knee and ankle torques with RMSE of 0.154 ± 0.06 and 0.106 ± 0.06 Nm/kg, respectively. Training the model on mode-specific data only produced significant reductions in stair descent, lowering knee and ankle RMSE by 0.06 ± 0.028 and 0.033 ± 0.008 Nm/kg respectively (p < 0.05). In addition, the TCN adapted to walking speed and slope shifts in reference commanded torque. These results demonstrate that this deep learning model not only removes the need for heuristic state machines and mode classification but can also reduce or remove the need for prosthesis assistance tuning entirely.

PMID:40012860 | PMC:PMC11864809 | DOI:10.1109/lra.2024.3374189

Microb Cell. 2025 Feb 20;12:1-8. doi: 10.15698/mic2025.02.841. eCollection 2025.

ABSTRACT

Molecular de-extinction has emerged as a novel strategy for studying biological molecules throughout evolutionary history. Among the myriad possibilities offered by ancient genomes and proteomes, antimicrobial peptides (AMPs) stand out as particularly promising alternatives to traditional antibiotics. Various strategies, including software tools and advanced deep learning models, have been used to mine these host defense peptides. For example, computational analysis of disulfide bond patterns has led to the identification of six previously uncharacterized β-defensins in extinct and critically endangered species. Additionally, artificial intelligence and machine learning have been utilized to uncover ancient antibiotics, revealing numerous candidates, including mammuthusin, and elephasin, which display inhibitory effects toward pathogens in vitro and in vivo. These innovations promise to discover novel antibiotics and deepen our insight into evolutionary processes.

PMID:40012704 | PMC:PMC11853161 | DOI:10.15698/mic2025.02.841

NPJ Biol Phys Mech. 2025;2(1):5. doi: 10.1038/s44341-025-00009-3. Epub 2025 Feb 24.

ABSTRACT

Biological systems are complex networks where measurable functions emerge from interactions among thousands of components. Many studies aim to link biological function with molecular elements, yet quantifying their contributions simultaneously remains challenging, especially at the single-cell level. We propose a machine-learning approach that integrates faceted data subsets to reconstruct a complete view of the system using conditional distributions. We develop both polynomial regression and neural network models, validated with two examples: a mechanical spring network under external forces and an 8-dimensional biological network involving the senescence marker P53, using single-cell data. Our results demonstrate successful system reconstruction from partial datasets, with predictive accuracy improving as more variables are measured. This approach offers a systematic method to integrate fragmented experimental data, enabling unbiased and holistic modeling of complex biological functions.

PMID:40012561 | PMC:PMC11850287 | DOI:10.1038/s44341-025-00009-3

J Cheminform. 2025 Feb 26;17(1):24. doi: 10.1186/s13321-025-00960-2.

ABSTRACT

In this study, we propose a neural network- based approach to analyze IR spectra and detect the presence of functional groups. Our neural network architecture is based on the concept of learning split representations. We demonstrate that our method achieves favorable validation performance using the NIST dataset. Furthermore, by incorporating additional data from the open-access research data repository Chemotion, we show that our model improves the classification performance for nitriles and amides. Scientific contribution: Our method exclusively uses IR data as input for a neural network, making its performance, unlike other well-performing models, independent of additional data types obtained from analytical measurements. Furthermore, our proposed method leverages a deep learning model that outperforms previous approaches, achieving F1 scores above 0.7 to identify 17 functional groups. By incorporating real-world data from various laboratories, we demonstrate how open-access, specialized research data repositories can serve as yet unexplored, valuable benchmark datasets for future machine learning research.

PMID:40011923 | DOI:10.1186/s13321-025-00960-2

BMC Geriatr. 2025 Feb 26;25(1):130. doi: 10.1186/s12877-025-05771-6.

ABSTRACT

BACKGROUND: Alzheimer's disease (AD) is a neurodegenerative disorder that significantly impacts health care worldwide, particularly among the elderly population. The accurate classification of AD stages is essential for slowing disease progression and guiding effective interventions. However, limited sample sizes continue to present a significant challenge in classifying the stages of AD progression. Addressing this obstacle is crucial for improving diagnostic accuracy and optimizing treatment strategies for those affected by AD.

METHODS: In this study, we proposed cross-scale equilibrium pyramid coupling (CSEPC), which is a novel diagnostic algorithm designed for small-sample multimodal medical imaging data. CSEPC leverages scale equilibrium theory and modal coupling properties to integrate semantic features from different imaging modalities and across multiple scales within each modality. The architecture first extracts balanced multiscale features from structural MRI (sMRI) data and functional MRI (fMRI) data using a cross-scale pyramid module. These features are then combined through a contrastive learning-based cosine similarity coupling mechanism to capture intermodality associations effectively. This approach enhances the representation of both inter- and intramodal features while significantly reducing the number of learning parameters, making it highly suitable for small sample environments. We validated the effectiveness of the CSEPC model through experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and demonstrated its superior performance in diagnosing and staging AD.

RESULTS: Our experimental results demonstrate that the proposed model matches or exceeds the performance of models used in previous studies in AD classification. Specifically, the model achieved an accuracy of 85.67% and an area under the curve (AUC) of 0.98 in classifying the progression from mild cognitive impairment (MCI) to AD. To further validate its effectiveness, we used our method to diagnose different stages of AD. In both classification tasks, our approach delivered superior performance.

CONCLUSIONS: In conclusion, the performance of our model in various tasks has demonstrated its significant potential in the field of small-sample multimodal medical imaging classification, particularly AD classification. This advancement could significantly assist clinicians in effectively managing and intervening in the disease progression of patients with early-stage AD.

PMID:40011826 | DOI:10.1186/s12877-025-05771-6

BMC Med Imaging. 2025 Feb 26;25(1):66. doi: 10.1186/s12880-025-01606-3.

ABSTRACT

BACKGROUND: This study employed a convolutional neural network (CNN) to analyze computed tomography (CT) scans with the aim of differentiating among renal tumors according to histologic sub-type.

METHODS: Contrast-enhanced CT images were collected from patients with renal tumors. The patient cohort was randomly split to create a training dataset (90%) and a testing dataset (10%). Following image dataset augmentation, Inception V3 and Resnet50 models were used to differentiate between renal tumors subtypes, including angiomyolipoma (AML), oncocytoma, clear cell renal cell carcinoma (ccRCC), chromophobe renal cell carcinoma (chRCC), and papillary renal cell carcinoma (pRCC). 5-fold cross validation was then used to evaluate the models in terms of classification performance.

RESULTS: The study cohort comprised 554 patients, including those with angiomyolipoma (n = 67), oncocytoma (n = 34), clear cell renal cell carcinoma (n = 246), chromophobe renal cell carcinoma (n = 124), and papillary renal cell carcinoma (n = 83). Dataset augmentation of the training dataset included this to 4238 CT images for analysis. The accuracy of the models was as follows: Inception V3 (0.830) and Resnet 50 (0.849).

CONCLUSION: This study demonstrated the efficacy of using deep learning models for the classification of renal tumor subtypes from contrast-enhanced CT images. While the models showed promising accuracy, further development is necessary to improve their clinical applicability.

PMID:40011809 | DOI:10.1186/s12880-025-01606-3