Insights Imaging. 2025 Jan 29;16(1):27. doi: 10.1186/s13244-025-01904-y.

ABSTRACT

OBJECTIVES: To develop and validate radiomics and deep learning models based on contrast-enhanced MRI (CE-MRI) for differentiating dual-phenotype hepatocellular carcinoma (DPHCC) from HCC and intrahepatic cholangiocarcinoma (ICC).

METHODS: Our study consisted of 381 patients from four centers with 138 HCCs, 122 DPHCCs, and 121 ICCs (244 for training and 62 for internal tests, centers 1 and 2; 75 for external tests, centers 3 and 4). Radiomics, deep transfer learning (DTL), and fusion models based on CE-MRI were established for differential diagnosis, respectively, and their diagnostic performances were compared using the confusion matrix and area under the receiver operating characteristic (ROC) curve (AUC).

RESULTS: The radiomics model demonstrated competent diagnostic performance, with a macro-AUC exceeding 0.9, and both accuracy and F1-score above 0.75 in the internal and external validation sets. Notably, the vgg19-combined model outperformed the radiomics and other DTL models. The fusion model based on vgg19 further improved diagnostic performance, achieving a macro-AUC of 0.990 (95% CI: 0.965-1.000), an accuracy of 0.935, and an F1-score of 0.937 in the internal test set. In the external test set, it similarly performed well, with a macro-AUC of 0.988 (95% CI: 0.964-1.000), accuracy of 0.875, and an F1-score of 0.885.

CONCLUSIONS: Both the radiomics and the DTL models were able to differentiate DPHCC from HCC and ICC before surgery. The fusion models showed better diagnostic accuracy, which has important value in clinical application.

CRITICAL RELEVANCE STATEMENT: MRI-based deep learning radiomics were able to differentiate DPHCC from HCC and ICC preoperatively, aiding clinicians in the identification and targeted treatment of these malignant hepatic tumors.

KEY POINTS: Fusion models may yield an incremental value over radiomics models in differential diagnosis. Radiomics and deep learning effectively differentiate the three types of malignant hepatic tumors. The fusion models may enhance clinical decision-making for malignant hepatic tumors.

PMID:39881111 | DOI:10.1186/s13244-025-01904-y

Insights Imaging. 2025 Jan 29;16(1):29. doi: 10.1186/s13244-025-01902-0.

ABSTRACT

OBJECTIVES: To determine whether deep learning-based reconstructions of zero-echo-time (ZTE-DL) sequences enhance image quality and bone visualization in cervical spine MRI compared to traditional zero-echo-time (ZTE) techniques, and to assess the added value of ZTE-DL sequences alongside standard cervical spine MRI for comprehensive pathology evaluation.

METHODS: In this retrospective study, 52 patients underwent cervical spine MRI using ZTE, ZTE-DL, and T2-weighted 3D sequences on a 1.5-Tesla scanner. ZTE-DL sequences were reconstructed from raw data using the AirReconDL algorithm. Three blinded readers independently evaluated image quality, artifacts, and bone delineation on a 5-point Likert scale. Cervical structures and pathologies, including soft tissue and bone components in spinal canal and neural foraminal stenosis, were analyzed. Image quality was quantitatively assessed by signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR).

RESULTS: Mean image quality scores were 2.0 ± 0.7 for ZTE and 3.2 ± 0.6 for ZTE-DL, with ZTE-DL exhibiting fewer artifacts and superior bone delineation. Significant differences were observed between T2-weighted and ZTE-DL sequences for evaluating intervertebral space, anterior osteophytes, spinal canal, and neural foraminal stenosis (p < 0.05), with ZTE-DL providing more accurate assessments. ZTE-DL also showed improved evaluation of the osseous components of neural foraminal stenosis compared to ZTE (p < 0.05).

CONCLUSIONS: ZTE-DL sequences offer superior image quality and bone visualization compared to ZTE sequences and enhance standard cervical spine MRI in assessing bone involvement in spinal canal and neural foraminal stenosis.

CRITICAL RELEVANCE STATEMENT: Deep learning-based reconstructions improve zero-echo-time sequences in cervical spine MRI by enhancing image quality and bone visualization. This advancement offers additional insights for assessing bone involvement in spinal canal and neural foraminal stenosis, advancing clinical radiology practice.

KEY POINTS: Conventional MRI encounters challenges with osseous structures due to low signal-to-noise ratio. Zero-echo-time (ZET) sequences offer CT-like images of the C-spine but with lower quality. Deep learning reconstructions improve image quality of zero-echo-time sequences. ZTE sequences with deep learning reconstructions refine cervical spine osseous pathology assessment. These sequences aid assessment of bone involvement in spinal and foraminal stenosis.

PMID:39881081 | DOI:10.1186/s13244-025-01902-0

Eur Radiol Exp. 2025 Jan 29;9(1):12. doi: 10.1186/s41747-025-00552-7.

ABSTRACT

BACKGROUND: Body composition scores allow for quantifying the volume and physical properties of specific tissues. However, their manual calculation is time-consuming and prone to human error. This study aims to develop and validate CompositIA, an automated, open-source pipeline for quantifying body composition scores from thoraco-abdominal computed tomography (CT) scans.

METHODS: A retrospective dataset of 205 contrast-enhanced thoraco-abdominal CT examinations was used for training, while 54 scans from a publicly available dataset were used for independent testing. Two radiology residents performed manual segmentation, identifying the centers of the L1 and L3 vertebrae and segmenting the corresponding axial slices. MultiResUNet was used to identify CT slices intersecting the L1 and L3 vertebrae, and its performance was evaluated using the mean absolute error (MAE). Two U-nets were used to segment the axial slices, with performance evaluated through the volumetric Dice similarity coefficient (vDSC). CompositIA's performance in quantifying body composition indices was assessed using mean percentage relative error (PRE), regression, and Bland-Altman analyses.

RESULTS: On the independent dataset, CompositIA achieved a MAE of about 5 mm in detecting slices intersecting the L1 and L3 vertebrae, with a MAE < 10 mm in at least 85% of cases and a vDSC greater than 0.85 in segmenting axial slices. Regression and Bland-Altman analyses demonstrated a strong linear relationship and good agreement between automated and manual scores (p values < 0.001 for all indices), with mean PREs ranging from 5.13% to 15.18%.

CONCLUSION: CompositIA facilitated the automated quantification of body composition scores, achieving high precision in independent testing.

RELEVANCE STATEMENT: CompositIA is an automated, open-source pipeline for quantifying body composition indices from CT scans, simplifying clinical assessments, and expanding their applicability.

KEY POINTS: Manual body composition assessment from CTs is time-consuming and prone to errors. CompositIA was trained on 205 CT scans and tested on 54 scans. CompositIA demonstrated mean percentage relative errors under 15% compared to manual indices. CompositIA simplifies body composition assessment through an artificial intelligence-driven and open-source pipeline.

PMID:39881078 | DOI:10.1186/s41747-025-00552-7

Appl Neuropsychol Adult. 2025 Jan 29:1-10. doi: 10.1080/23279095.2025.2457999. Online ahead of print.

ABSTRACT

In the fields of engineering, science, technology, and medicine, artificial intelligence (AI) has made significant advancements. In particular, the application of AI techniques in medicine, such as machine learning (ML) and deep learning (DL), is rapidly growing and offers great potential for aiding physicians in the early diagnosis of illnesses. Depression, one of the most prevalent and debilitating mental illnesses, is projected to become the leading cause of disability worldwide by 2040. For early diagnosis, a patient-friendly, cost-effective approach based on readily observable and objective indicators is essential. The objective of this research is to develop machine learning and deep learning techniques that utilize electroencephalogram (EEG) signals to diagnose depression. Different statistical features were extracted from the EEG signals and fed into the models. Three classifiers were constructed: 1D Convolutional Neural Network (1DCNN), Support Vector Machine (SVM), and Logistic Regression (LR). The methods were tested on a dataset comprising EEG signals from 34 patients with Major Depressive Disorder (MDD) and 30 healthy subjects. The signals were collected under three distinct conditions: TASK, when the subject was performing a task; Eye Close (EC), when the subject's eyes were closed; and Eye Open (EO), when the subject's eyes were open. All three classifiers were applied to each of the three types of signals, resulting in nine (3 × 3) experiments. The results showed that TASK signals yielded the highest accuracies of 88.4%, 89.3%, and 90.21% for LR, SVM, and 1DCNN, respectively, compared to EC and EO signals. Additionally, the proposed methods outperformed some state-of-the-art approaches. These findings highlight the potential of EEG-based approaches for the clinical diagnosis of depression and provide promising avenues for further research. Additionally, the proposed methodology demonstrated statistically significant improvements in classification accuracy, with p-values < 0.05, ensuring robustness and reliability.

PMID:39879638 | DOI:10.1080/23279095.2025.2457999

J Med Internet Res. 2025 Jan 29;27:e57723. doi: 10.2196/57723.

ABSTRACT

BACKGROUND: Neuroimaging segmentation is increasingly important for diagnosing and planning treatments for neurological diseases. Manual segmentation is time-consuming, apart from being prone to human error and variability. Transformers are a promising deep learning approach for automated medical image segmentation.

OBJECTIVE: This scoping review will synthesize current literature and assess the use of various transformer models for neuroimaging segmentation.

METHODS: A systematic search in major databases, including Scopus, IEEE Xplore, PubMed, and ACM Digital Library, was carried out for studies applying transformers to neuroimaging segmentation problems from 2019 through 2023. The inclusion criteria allow only for peer-reviewed journal papers and conference papers focused on transformer-based segmentation of human brain imaging data. Excluded are the studies dealing with nonneuroimaging data or raw brain signals and electroencephalogram data. Data extraction was performed to identify key study details, including image modalities, datasets, neurological conditions, transformer models, and evaluation metrics. Results were synthesized using a narrative approach.

RESULTS: Of the 1246 publications identified, 67 (5.38%) met the inclusion criteria. Half of all included studies were published in 2022, and more than two-thirds used transformers for segmenting brain tumors. The most common imaging modality was magnetic resonance imaging (n=59, 88.06%), while the most frequently used dataset was brain tumor segmentation dataset (n=39, 58.21%). 3D transformer models (n=42, 62.69%) were more prevalent than their 2D counterparts. The most developed were those of hybrid convolutional neural network-transformer architectures (n=57, 85.07%), where the vision transformer is the most frequently used type of transformer (n=37, 55.22%). The most frequent evaluation metric was the Dice score (n=63, 94.03%). Studies generally reported increased segmentation accuracy and the ability to model both local and global features in brain images.

CONCLUSIONS: This review represents the recent increase in the adoption of transformers for neuroimaging segmentation, particularly for brain tumor detection. Currently, hybrid convolutional neural network-transformer architectures achieve state-of-the-art performances on benchmark datasets over standalone models. Nevertheless, their applicability remains highly limited by high computational costs and potential overfitting on small datasets. The heavy reliance of the field on the brain tumor segmentation dataset hints at the use of a more diverse set of datasets to validate the performances of models on a variety of neurological diseases. Further research is needed to define the optimal transformer architectures and training methods for clinical applications. Continuing development may make transformers the state-of-the-art for fast, accurate, and reliable brain magnetic resonance imaging segmentation, which could lead to improved clinical tools for diagnosing and evaluating neurological disorders.

PMID:39879621 | DOI:10.2196/57723

JMIR Public Health Surveill. 2025 Jan 29;11:e63809. doi: 10.2196/63809.

ABSTRACT

BACKGROUND: Suicide represents a critical public health concern, and machine learning (ML) models offer the potential for identifying at-risk individuals. Recent studies using benchmark datasets and real-world social media data have demonstrated the capability of pretrained large language models in predicting suicidal ideation and behaviors (SIB) in speech and text.

OBJECTIVE: This study aimed to (1) develop and implement ML methods for predicting SIBs in a real-world crisis helpline dataset, using transformer-based pretrained models as a foundation; (2) evaluate, cross-validate, and benchmark the model against traditional text classification approaches; and (3) train an explainable model to highlight relevant risk-associated features.

METHODS: We analyzed chat protocols from adolescents and young adults (aged 14-25 years) seeking assistance from a German crisis helpline. An ML model was developed using a transformer-based language model architecture with pretrained weights and long short-term memory layers. The model predicted suicidal ideation (SI) and advanced suicidal engagement (ASE), as indicated by composite Columbia-Suicide Severity Rating Scale scores. We compared model performance against a classical word-vector-based ML model. We subsequently computed discrimination, calibration, clinical utility, and explainability information using a Shapley Additive Explanations value-based post hoc estimation model.

RESULTS: The dataset comprised 1348 help-seeking encounters (1011 for training and 337 for testing). The transformer-based classifier achieved a macroaveraged area under the curve (AUC) receiver operating characteristic (ROC) of 0.89 (95% CI 0.81-0.91) and an overall accuracy of 0.79 (95% CI 0.73-0.99). This performance surpassed the word-vector-based baseline model (AUC-ROC=0.77, 95% CI 0.64-0.90; accuracy=0.61, 95% CI 0.61-0.80). The transformer model demonstrated excellent prediction for nonsuicidal sessions (AUC-ROC=0.96, 95% CI 0.96-0.99) and good prediction for SI and ASE, with AUC-ROCs of 0.85 (95% CI 0.97-0.86) and 0.87 (95% CI 0.81-0.88), respectively. The Brier Skill Score indicated a 44% improvement in classification performance over the baseline model. The Shapley Additive Explanations model identified language features predictive of SIBs, including self-reference, negation, expressions of low self-esteem, and absolutist language.

CONCLUSIONS: Neural networks using large language model-based transfer learning can accurately identify SI and ASE. The post hoc explainer model revealed language features associated with SI and ASE. Such models may potentially support clinical decision-making in suicide prevention services. Future research should explore multimodal input features and temporal aspects of suicide risk.

PMID:39879608 | DOI:10.2196/63809

JMIR Form Res. 2025 Jan 29;9:e66330. doi: 10.2196/66330.

ABSTRACT

BACKGROUND: Estimating the prevalence of schizophrenia in the general population remains a challenge worldwide, as well as in Japan. Few studies have estimated schizophrenia prevalence in the Japanese population and have often relied on reports from hospitals and self-reported physician diagnoses or typical schizophrenia symptoms. These approaches are likely to underestimate the true prevalence owing to stigma, poor insight, or lack of access to health care among respondents. To address these issues, we previously developed an artificial neural network (ANN)-based schizophrenia classification model (SZ classifier) using data from a large-scale Japanese web-based survey to enhance the comprehensiveness of schizophrenia case identification in the general population. In addition, we also plan to introduce a population-based survey to collect general information and sample participants matching the population's demographic structure, thereby achieving a precise estimate of the prevalence of schizophrenia in Japan.

OBJECTIVE: This study aimed to estimate the prevalence of schizophrenia by applying the SZ classifier to random samples from the Japanese population.

METHODS: We randomly selected a sample of 750 participants where the age, sex, and regional distributions were similar to Japan's demographic structure from a large-scale Japanese web-based survey. Demographic data, health-related backgrounds, physical comorbidities, psychiatric comorbidities, and social comorbidities were collected and applied to the SZ classifier, as this information was also used for developing the SZ classifier. The crude prevalence of schizophrenia was calculated through the proportion of positive cases detected by the SZ classifier. The crude estimate was further refined by excluding false-positive cases and including false-negative cases to determine the actual prevalence of schizophrenia.

RESULTS: Out of 750 participants, 62 were classified as schizophrenia cases by the SZ classifier, resulting in a crude prevalence of schizophrenia in the general population of Japan of 8.3% (95% CI 6.6%-10.1%). Among these 62 cases, 53 were presumed to be false positives, and 3 were presumed to be false negatives. After adjustment, the actual prevalence of schizophrenia in the general population was estimated to be 1.6% (95% CI 0.7%-2.5%).

CONCLUSIONS: This estimated prevalence was slightly higher than that reported in previous studies, possibly due to a more comprehensive disease classification methodology or, conversely, model limitations. This study demonstrates the capability of an ANN-based model to improve the estimation of schizophrenia prevalence in the general population, offering a novel approach to public health analysis.

PMID:39879582 | DOI:10.2196/66330

Brief Bioinform. 2024 Nov 22;26(1):bbaf039. doi: 10.1093/bib/bbaf039.

ABSTRACT

Pathway analysis plays a critical role in bioinformatics, enabling researchers to identify biological pathways associated with various conditions by analyzing gene expression data. However, the rise of large, multi-center datasets has highlighted limitations in traditional methods like Over-Representation Analysis (ORA) and Functional Class Scoring (FCS), which struggle with low signal-to-noise ratios (SNR) and large sample sizes. To tackle these challenges, we use a deep learning-based classification method, Gene PointNet, and a novel $P$-value computation approach leveraging the confusion matrix to address pathway analysis tasks. We validated our method effectiveness through a comparative study using a simulated dataset and RNA-Seq data from The Cancer Genome Atlas breast cancer dataset. Our method was benchmarked against traditional techniques (ORA, FCS), shallow machine learning models (logistic regression, support vector machine), and deep learning approaches (DeepHisCom, PASNet). The results demonstrate that GPNet outperforms these methods in low-SNR, large-sample datasets, where it remains robust and reliable, significantly reducing both Type I error and improving power. This makes our method well suited for pathway analysis in large, multi-center studies. The code can be found at https://github.com/haolu123/GPNet_pathway">https://github.com/haolu123/GPNet_pathway.

PMID:39879387 | DOI:10.1093/bib/bbaf039

Brief Bioinform. 2024 Nov 22;26(1):bbaf037. doi: 10.1093/bib/bbaf037.

ABSTRACT

This study aimed to investigate the genetic association between glioblastoma (GBM) and unsupervised deep learning-derived imaging phenotypes (UDIPs). We employed a combination of genome-wide association study (GWAS) data, single-nucleus RNA sequencing (snRNA-seq), and scPagwas (pathway-based polygenic regression framework) methods to explore the genetic links between UDIPs and GBM. Two-sample Mendelian randomization analyses were conducted to identify causal relationships between UDIPs and GBM. Colocalization analysis was performed to validate genetic associations, while scPagwas analysis was used to evaluate the relevance of key UDIPs to GBM at the cellular level. Among 512 UDIPs tested, 23 were found to have significant causal associations with GBM. Notably, UDIPs such as T1-33 (OR = 1.007, 95% CI = 1.001 to 1.012, P = .022), T1-34 (OR = 1.012, 95% CI = 1.001-1.023, P = .028), and T1-96 (OR = 1.009, 95% CI = 1.001-1.019, P = .046) were found to have a genetic association with GBM. Furthermore, T1-34 and T1-96 were significantly associated with GBM recurrence, with P-values < .0001 and P < .001, respectively. In addition, scPagwas analysis revealed that T1-33, T1-34, and T1-96 are distinctively linked to different GBM subtypes, with T1-33 showing strong associations with the neural progenitor-like subtype (NPC2), T1-34 with mesenchymal (MES2) and neural progenitor (NPC1) cells, and T1-96 with the NPC2 subtype. T1-33, T1-34, and T1-96 hold significant potential for predicting tumor recurrence and aiding in the development of personalized GBM treatment strategies.

PMID:39879386 | DOI:10.1093/bib/bbaf037

PLoS One. 2025 Jan 29;20(1):e0316493. doi: 10.1371/journal.pone.0316493. eCollection 2025.

ABSTRACT

The emergence of Next Generation Sequencing (NGS) technology has catalyzed a paradigm shift in clinical diagnostics and personalized medicine, enabling unprecedented access to high-throughput microbiome data. However, the inherent high dimensionality, noise, and variability of microbiome data present substantial obstacles to conventional statistical methods and machine learning techniques. Even the promising deep learning (DL) methods are not immune to these challenges. This paper introduces a novel feature engineering method that circumvents these limitations by amalgamating two feature sets derived from input data to generate a new dataset, which is then subjected to feature selection. This innovative approach markedly enhances the Area Under the Curve (AUC) performance of the Deep Neural Network (DNN) algorithm in colorectal cancer (CRC) detection using gut microbiome data, elevating it from 0.800 to 0.923. The proposed method constitutes a significant advancement in the field, providing a robust solution to the intricacies of microbiome data analysis and amplifying the potential of DL methods in disease detection.

PMID:39879257 | DOI:10.1371/journal.pone.0316493

Eur J Nucl Med Mol Imaging. 2025 Jan 29. doi: 10.1007/s00259-025-07091-8. Online ahead of print.

ABSTRACT

PURPOSE: Advancements of deep learning in medical imaging are often constrained by the limited availability of large, annotated datasets, resulting in underperforming models when deployed under real-world conditions. This study investigated a generative artificial intelligence (AI) approach to create synthetic medical images taking the example of bone scintigraphy scans, to increase the data diversity of small-scale datasets for more effective model training and improved generalization.

METHODS: We trained a generative model on 99mTc-bone scintigraphy scans from 9,170 patients in one center to generate high-quality and fully anonymized annotated scans of patients representing two distinct disease patterns: abnormal uptake indicative of (i) bone metastases and (ii) cardiac uptake indicative of cardiac amyloidosis. A blinded reader study was performed to assess the clinical validity and quality of the generated data. We investigated the added value of the generated data by augmenting an independent small single-center dataset with synthetic data and by training a deep learning model to detect abnormal uptake in a downstream classification task. We tested this model on 7,472 scans from 6,448 patients across four external sites in a cross-tracer and cross-scanner setting and associated the resulting model predictions with clinical outcomes.

RESULTS: The clinical value and high quality of the synthetic imaging data were confirmed by four readers, who were unable to distinguish synthetic scans from real scans (average accuracy: 0.48% [95% CI 0.46-0.51]), disagreeing in 239 (60%) of 400 cases (Fleiss' kappa: 0.18). Adding synthetic data to the training set improved model performance by a mean (± SD) of 33(± 10)% AUC (p < 0.0001) for detecting abnormal uptake indicative of bone metastases and by 5(± 4)% AUC (p < 0.0001) for detecting uptake indicative of cardiac amyloidosis across both internal and external testing cohorts, compared to models without synthetic training data. Patients with predicted abnormal uptake had adverse clinical outcomes (log-rank: p < 0.0001).

CONCLUSIONS: Generative AI enables the targeted generation of bone scintigraphy images representing different clinical conditions. Our findings point to the potential of synthetic data to overcome challenges in data sharing and in developing reliable and prognostic deep learning models in data-limited environments.

PMID:39878897 | DOI:10.1007/s00259-025-07091-8

Br J Radiol. 2025 Jan 29:tqaf018. doi: 10.1093/bjr/tqaf018. Online ahead of print.

ABSTRACT

In a rapidly evolving healthcare environment, artificial intelligence (AI) is transforming diagnostic techniques and personalised medicine. This is also seen in osseous biopsies. AI applications in radiomics, histopathology, predictive modelling, biopsy navigation, and interdisciplinary communication are reshaping how bone biopsies are conducted and interpreted. We provide a brief review of AI in image- guided biopsy of bone tumours (primary and secondary) and specimen handling, in the era of personalised medicine. This paper explores AI's role in enhancing diagnostic accuracy, improving safety in biopsies, and enabling more precise targeting in bone lesion biopsies, ultimately contributing to better patient outcomes in personalised medicine. We dive into various AI technologies applied to osseous biopsies, such as traditional machine learning, deep learning, radiomics, simulation and generative models. We explore their roles in tumour board meetings, communication between clinicians, radiologists, and pathologists. Additionally, we inspect ethical considerations associated with the integration of AI in bone biopsy procedures, technical limitations, and we delve into health equity, generalisability, deployment issues, and reimbursement challenges in AI-powered healthcare. Finally, we explore potential future developments and offer a list of open-source AI tools and algorithms relevant to bone biopsies, which we include to encourage further discussion and research.

PMID:39878877 | DOI:10.1093/bjr/tqaf018

Clin Oral Investig. 2025 Jan 29;29(2):97. doi: 10.1007/s00784-025-06183-x.

ABSTRACT

OBJECTIVES: To evaluate recent advances in the automatic multimodal registration of cone-beam computed tomography (CBCT) and intraoral scans (IOS) and their clinical significance in dentistry.

METHODS: A comprehensive literature search was conducted in October 2024 across the PubMed, Web of Science, and IEEE Xplore databases, including studies that were published in the past decade. The inclusion criteria were as follows: English-language studies, randomized and nonrandomized controlled trials, cohort studies, case-control studies, cross-sectional studies, and retrospective studies.

RESULTS: Of the 493 articles identified, 22 met the inclusion criteria. Among these, 14 studies used geometry-based methods, 7 used artificial intelligence (AI) techniques, and 1 compared the accuracy of both approaches. Geometry-based methods primarily utilize two-stage coarse-to-fine registration algorithms, which require relatively fewer computational resources. In contrast, AI methods leverage advanced deep learning models, achieving significant improvements in automation and robustness.

CONCLUSIONS: Recent advances in CBCT and IOS registration technologies have considerably increased the efficiency and accuracy of 3D dental modelling, and these technologies show promise for application in orthodontics, implantology, and oral surgery. Geometry-based algorithms deliver reliable performance with low computational demand, whereas AI-driven approaches demonstrate significant potential for achieving fully automated and highly accurate registration. Future research should focus on challenges such as unstable registration landmarks or limited dataset diversity, to ensure their stability in complex clinical scenarios.

PMID:39878846 | DOI:10.1007/s00784-025-06183-x

Bioresour Bioprocess. 2025 Jan 29;12(1):7. doi: 10.1186/s40643-024-00835-8.

ABSTRACT

Feruloyl esterases (FEs, EC 3.1.1.73) play a crucial role in biological synthesis and metabolism. However, the identification of versatile FEs, capable of catalyzing a wide range of substrates, remains a challenge. In this study, we obtained 2085 FE sequences from the BRENDA database and initiated with an enzyme similarity network analysis, revealing three main clusters (1-3). Notably, both cluster 1 and cluster 3 included the characterized FEs, which exhibited significant differences in sequence length. Subsequent phylogenetic analysis of these clusters unveiled a correlation between phylogenetic classification and substrate promiscuity, and enzymes with broad substrate scope tended to locate within specific branches of the phylogenetic tree. Further, molecular dynamics simulations and dynamical cross-correlation matrix analysis were employed to explore structural dynamics differences between promiscuous and substrate-specific FEs. Finally, to expand the repertoire of versatile FEs, we employed deep learning models to predict potentially promiscuous enzymes and identified 38 and 75 potential versatile FEs from cluster 1 and cluster 3 with a probability score exceeding 90%. Our findings underscore the utility of integrating phylogenetic and structural features with deep learning approaches for mining versatile FEs, shedding light on unexplored enzymatic diversity and expanding the repertoire of biocatalysts for synthetic applications.

PMID:39878830 | DOI:10.1186/s40643-024-00835-8

Physiol Rep. 2025 Feb;13(3):e70187. doi: 10.14814/phy2.70187.

ABSTRACT

The maintenance of an appropriate ratio of body fat to muscle mass is essential for the preservation of health and performance, as excessive body fat is associated with an increased risk of various diseases. Accurate body composition assessment requires precise segmentation of structures. In this study we developed a novel automatic machine learning approach for volumetric segmentation and quantitative assessment of MRI volumes and investigated the efficacy of using a machine learning algorithm to assess muscle, subcutaneous adipose tissue (SAT), and bone volume of the thigh before and after a strength training. Eighteen healthy, young, female volunteers were randomly allocated to two groups: intervention group (IG) and control group (CG). The IG group followed an 8-week strength endurance training plan that was conducted two times per week. Before and after the training, the subjects of both groups underwent MRI scanning. The evaluation of the image data was performed by a machine learning system which is based on a 3D U-Net-based Convolutional Neural Network. The volumes of muscle, bone, and SAT were each examined using a 2 (GROUP [IG vs. CG]) × 2 (TIME [pre-intervention vs. post-intervention]) analysis of variance (ANOVA) with repeated measures for the factor TIME. The results of the ANOVA demonstrate significant TIME × GROUP interaction effects for the muscle volume (F1,16 = 12.80, p = 0.003, ηP 2 = 0.44) with an increase of 2.93% in the IG group and no change in the CG (-0.62%, p = 0.893). There were no significant changes in bone or SAT volume between the groups. This study supports the use of artificial intelligence systems to analyze MRI images as a reliable tool for monitoring training responses on body composition.

PMID:39878619 | DOI:10.14814/phy2.70187

Med Phys. 2025 Jan 29. doi: 10.1002/mp.17642. Online ahead of print.

ABSTRACT

BACKGROUND: Computed tomography angiography (CTA) is used to screen for coronary artery calcification. As the coronary artery has complicated structure and tiny lumen, manual screening is a time-consuming task. Recently, many deep learning methods have been proposed for the segmentation (SEG) of coronary artery and calcification, however, they often neglect leveraging related anatomical prior knowledge, resulting in low accuracy and instability.

PURPOSE: This study aims to build a deep learning based SEG framework, which leverages anatomical prior knowledge of coronary artery and calcification, to improve the SEG accuracy. Moreover, based on the SEG results, this study also try to reveal the predictive ability of the volume ratio of coronary artery and calcification for rotational atherectomy (RA).

METHODS: We present a new SEG framework, which is composed of four modules: the variational autoencoder based centerline extraction (CE) module, the self-attention (SA) module, the logic operation (LO) module, and the SEG module. Specifically, the CE module is used to crop a series of 3D CTA patches along the coronary artery, from which the continuous property of vessels can be utilized by the SA module to produce vessel-related features. According to the spatial relations between coronary artery lumen and calcification regions, the LO module with logic union and intersection is designed to refine the vessel-related features into lumen- and calcification-related features, based on which SEG results can be produced by the SEG module.

RESULTS: Experimental results demonstrate that our framework outperforms the state-of-the-art methods on CTA image dataset of 72 patients with statistical significance. Ablation experiments confirm that the proposed modules have positive impacts to the SEG results. Moreover, based on the volume ratio of segmented coronary artery and calcification, the prediction accuracy of RA is 0.75.

CONCLUSIONS: Integrating anatomical prior knowledge of coronary artery and calcification into the deep learning based SEG framework can effectively enhance the performance. Moreover, the volume ratio of segmented coronary artery and calcification is a good predictive factor for RA.

PMID:39878608 | DOI:10.1002/mp.17642

Med Phys. 2025 Jan 29. doi: 10.1002/mp.17635. Online ahead of print.

ABSTRACT

BACKGROUND: Stereotactic radiosurgery (SRS) is widely used for managing brain metastases (BMs), but an adverse effect, radionecrosis, complicates post-SRS management. Differentiating radionecrosis from tumor recurrence non-invasively remains a major clinical challenge, as conventional imaging techniques often necessitate surgical biopsy for accurate diagnosis. Machine learning and deep learning models have shown potential in distinguishing radionecrosis from tumor recurrence. However, their clinical adoption is hindered by a lack of explainability, limiting understanding and trust in their diagnostic decisions.

PURPOSE: To utilize a novel neural ordinary differential equation (NODE) model for discerning BM post-SRS radionecrosis from recurrence. This approach integrates image-deep features, genomic biomarkers, and non-image clinical parameters within a synthesized latent feature space. The trajectory of each data sample towards the diagnosis decision can be visualized within this feature space, offering a new angle on radiogenomic data analysis foundational for AI explainability.

METHODS: By hypothesizing that deep feature extraction can be modeled as a spatiotemporally continuous process, we designed a novel model based on heavy ball NODE (HBNODE) in which deep feature extraction was governed by a second-order ODE. This approach enabled tracking of deep neural network (DNN) behavior by solving the HBNODE and observing the stepwise derivative evolution. Consequently, the trajectory of each sample within the Image-Genomic-Clinical (I-G-C) space became traceable. A decision-making field (F) was reconstructed within the feature space, with its gradient vectors directing the data samples' trajectories and intensities showing the potential. The evolution of F reflected the cumulative feature contributions at intermediate states to the final diagnosis, enabling quantitative and dynamic comparisons of the relative contribution of each feature category over time. A velocity curve was designed to determine key intermediate states (locoregional ∇F = 0) that are most predictive. Subsequently, a non-parametric model aggregated the optimal solutions from these key states to predict outcomes. Our dataset included 90 BMs from 62 NSCLC patients, and 3-month post-SRS T1+c MR image features, seven NSCLC genomic features, and seven clinical features were analyzed. An 8:2 train/test assignment was employed, and five independent models were trained to ensure robustness. Performance was benchmarked in sensitivity, specificity, accuracy, and ROCAUC, and results were compared against (1) a DNN using only image-based features, and (2) a combined "I+G+C" features without the HBNODE model.

RESULTS: The temporal evolution of gradient vectors and potential fields in F suggested that clinical features contribute the most during the initial stages of the HBNODE implementation, followed by imagery features taking dominance in the latter ones, while genomic features contribute the least throughout the process. The HBNODE model successfully identified and assembled key intermediate states, exhibiting competitive performance with an ROCAUC of 0.88 ± 0.04, sensitivity of 0.79 ± 0.02, specificity of 0.86 ± 0.01, and accuracy of 0.84 ± 0.01, where the uncertainties represent standard deviations. For comparison, the image-only DNN model achieved an ROCAUC of 0.71 ± 0.05 and sensitivity of 0.66 ± 0.32 (p = 0.086), while the "I+G+C" model without HBNODE reported an ROCAUC of 0.81 ± 0.02 and sensitivity of 0.58 ± 0.11 (p = 0.091).

CONCLUSION: The HBNODE model effectively identifies BM radionecrosis from recurrence, enhancing explainability within XAI frameworks. Its performance encourages further exploration in clinical settings and suggests potential applicability across various XAI domains.

PMID:39878595 | DOI:10.1002/mp.17635

J Chem Inf Model. 2025 Jan 29. doi: 10.1021/acs.jcim.4c01338. Online ahead of print.

ABSTRACT

Cell-penetrating peptides (CPPs) are short peptides capable of penetrating cell membranes, making them valuable for drug delivery and intracellular targeting. Accurate prediction of CPPs can streamline experimental validation in the lab. This study aims to assess pretrained protein language models (pLMs) for their effectiveness in representing CPPs and develop a reliable model for CPP classification. We evaluated peptide embeddings generated from BEPLER, CPCProt, SeqVec, various ESM variants (ESM, ESM-2 with expanded feature set, ESM-1b, and ESM-1v), ProtT5-XL UniRef50, ProtT5-XL BFD, and ProtBERT. We developed pLM4CCPs, a novel deep learning architecture using convolutional neural networks (CNNs) as the classifier for binary classification of CPPs. pLM4CCPs demonstrated superior performance over existing state-of-the-art CPP prediction models, achieving improvements in accuracy (ACC) by 4.9-5.5%, Matthews correlation coefficient (MCC) by 9.3-10.2%, and sensitivity (Sn) by 14.1-19.6%. Among all the tested models, ESM-1280 and ProtT5-XL BFD demonstrated the highest overall performance on the kelm data set. ESM-1280 achieved an ACC of 0.896, an MCC of 0.796, a Sn of 0.844, and a specificity (Sp) of 0.978. ProtT5-XL BFD exhibited superior performance with an ACC of 0.901, an MCC of 0.802, an Sn of 0.885, and an Sp of 0.917. pLM4CCPs combine predictions from multiple models to provide a consensus on whether a given peptide sequence is classified as a CPP or non-CPP. This approach will enhance prediction reliability by leveraging the strengths of each individual model. A user-friendly web server for bioactivity predictions, along with data sets, is available at https://ry2acnp6ep.us-east-1.awsapprunner.com. The source code and protocol for adapting pLM4CPPs can be accessed on GitHub at https://github.com/drkumarnandan/pLM4CPPs. This platform aims to advance CPP prediction and peptide functionality modeling, aiding researchers in exploring peptide functionality effectively.

PMID:39878455 | DOI:10.1021/acs.jcim.4c01338

Int J Gynecol Cancer. 2025 Jan;35(1):100017. doi: 10.1016/j.ijgc.2024.100017. Epub 2024 Dec 17.

ABSTRACT

OBJECTIVE: Evaluation of prognostic factors is crucial in patients with endometrial cancer for optimal treatment planning and prognosis assessment. This study proposes a deep learning pipeline for tumor and uterus segmentation from magnetic resonance imaging (MRI) images to predict deep myometrial invasion and cervical stroma invasion and thus assist clinicians in pre-operative workups.

METHODS: Two experts consensually reviewed the MRIs and assessed myometrial invasion and cervical stromal invasion as per the International Federation of Gynecology and Obstetrics staging classification, to compare the diagnostic performance of the model with the radiologic consensus. The deep learning method was trained using sagittal T2-weighted images from 142 patients and tested with a 3-fold stratified test with 36 patients in each fold. Our solution is based on a segmentation module, which employed a 2-stage pipeline for efficient uterus in the whole MRI volume and then tumor segmentation in the uterus predicted region of interest.

RESULTS: A total of 178 patients were included. For deep myometrial invasion prediction, the model achieved an average balanced test accuracy over 3-folds of 0.702, while experts reached an average accuracy of 0.769. For cervical stroma invasion prediction, our model demonstrated an average balanced accuracy of 0.721 on the 3-fold test set, while experts achieved an average balanced accuracy of 0.859. Additionally, the accuracy rates for uterus and tumor segmentation, measured by the Dice score, were 0.847 and 0.579 respectively.

CONCLUSION: Despite the current challenges posed by variations in data, class imbalance, and the presence of artifacts, our fully automatic approach holds great promise in supporting in pre-operative staging. Moreover, it demonstrated a robust capability to segment key regions of interest, specifically the uterus and tumors, highlighting the positive impact our solution can bring to health care imaging.

PMID:39878275 | DOI:10.1016/j.ijgc.2024.100017

ACS Electrochem. 2024 Oct 3;1(1):52-62. doi: 10.1021/acselectrochem.4c00014. eCollection 2025 Jan 2.

ABSTRACT

In electrochemical analysis, mechanism assignment is fundamental to understanding the chemistry of a system. The detection and classification of electrochemical mechanisms in cyclic voltammetry set the foundation for subsequent quantitative evaluation and practical application, but are often based on relatively subjective visual analyses. Deep-learning (DL) techniques provide an alternative, automated means that can support experimentalists in mechanism assignment. Herein, we present a custom DL architecture dubbed as EchemNet, capable of assigning both voltage windows and mechanism classes to electrochemical events within cyclic voltammograms of multiple redox events. The developed technique detects over 96% of all electrochemical events in simulated test data and shows a classification accuracy of up to 97.2% on redox events with 8 known mechanisms. This newly developed DL model, the first of its kind, proves the feasibility of redox-event detection and electrochemical mechanism classification with minimal a priori knowledge. The DL model will augment human researchers' productivity and constitute a critical component in a general-purpose autonomous electrochemistry laboratory.

PMID:39878149 | PMC:PMC11728721 | DOI:10.1021/acselectrochem.4c00014