Sci Rep. 2025 Jul 19;15(1):26237. doi: 10.1038/s41598-025-10629-3.

ABSTRACT

The critical role of gut microbiota in human health and disease has been increasingly illustrated over the past decades, with a significant amount of research demonstrating an unmet need for self-monitor of the fecal microbial composition in an easily-accessible, rapid-time manner. In this study, we employed a tool for Smartphone Microbiome Evaluation and Analysis in Rapid-time (SMEAR) that uses images of fecal smears to predict microbial compositional characteristics in a human cohort. A subset of human fecal samples was randomly retrieved from the second wave of data collection in the Healthy Life in an Urban Setting (HELIUS) study cohort. Per sample, 16S rRNA gene sequencing data was generated in addition to an image of a fecal smear, spread on a standard A4 paper. Metagenomics-paired pictures were used to validate a computer vision-based technology to classify whether the sample is of low or high relative abundance of the 50 most abundant genera, and α-diversity (Shannon-index). In total, 888 fecal samples were used as an application of the SMEAR technology. SMEAR gave accurate predictions whether a fecal sample is of low or high relative abundance of Sporobacter, Oscillibacter and Intestinimonas (very good performance, AUC > 0.8, p-value < 0.001, for all models), as well as Neglecta, Megasphaera, Lachnospira, Methanobrevibacter, Harryflintia, Roseburia, and Dialister (good performance, AUC > 0.75, p-value < 0.001, for all models). Likewise, SMEAR could classify whether a fecal sample was of low or high α-diversity (AUC = 0.83, p-value < 0.001). Our study demonstrates that SMEAR robustly predicts microbial composition and diversity from digital images of fecal smears in a human cohort. These findings establish SMEAR as a new benchmark for rapid, cost-effective microbiome diagnostics and pave the way for its direct application in research settings and clinical validation.

PMID:40683926 | DOI:10.1038/s41598-025-10629-3

Sci Rep. 2025 Jul 19;15(1):26222. doi: 10.1038/s41598-025-11775-4.

NO ABSTRACT

PMID:40683924 | DOI:10.1038/s41598-025-11775-4

Sci Rep. 2025 Jul 19;15(1):26262. doi: 10.1038/s41598-025-10804-6.

ABSTRACT

To enhance thrombolysis eligibility in acute ischemic stroke, we developed a deep learning model to estimate stroke onset within 4.5 h using diffusion-weighted imaging (DWI) and fluid-attenuated inversion recovery (FLAIR) images. Given the variability in human interpretation, our multimodal Res-U-Net (mRUNet) model integrates a modified U-Net and ResNet-34 to classify stroke onset as < 4.5 or ≥ 4.5 h. Using DWI and FLAIR images from patients scanned within 24 h of symptom onset, the modified U-Net generated a DWI-FLAIR mismatch image, while ResNet-34 performed the final classification. mRUNet was evaluated against ResNet-34 and DenseNet-121 on an internal test set (n = 123) and two external test sets: a single-center (n = 468) and a multi-center (n = 1151). mRUNet achieved an area under the receiver operating characteristic curve (AUC-ROC) of 0.903 on the internal set and 0.910 and 0.868 on external sets, significantly outperforming ResNet-34 and DenseNet-121. Our mRUNet model demonstrated robust and consistent classification of the 4.5-h onset-time window across datasets. By leveraging DWI and FLAIR images as a tissue clock, this model may support timely and individualized thrombolysis in patients with unclear stroke onset, such as those with wake-up stroke, in clinical settings.

PMID:40683923 | DOI:10.1038/s41598-025-10804-6

Sci Rep. 2025 Jul 19;15(1):26251. doi: 10.1038/s41598-025-09155-z.

ABSTRACT

Predicting the Remaining Useful Life (RUL) of turbofan engines can prevent air disasters caused by component degradation. It is an important procedure in prognostics and health management (PHM). Therefore, a deep learning-based RUL prediction approach is proposed. The CMAPSS benchmark dataset is used to determine the RUL of aviation engines, focusing specifically on the FD001 and FD003 sub-datasets.In this study, we propose a CAELSTM (Convolutional Autoencoder and Attention-based LSTM) hybrid model for RUL prediction. First, the sub-datasets are preprocessed, and a piecewise linear degradation model is applied. The proposed model utilizes an autoencoder followed by an LSTM layer with an attention mechanism, which focuses on the most relevant components of the sequences. A fully connected layer of the convolutional neural network is used to further process the important features. Finally, the proposed model is evaluated and compared with other approaches. The results show that the model surpasses state-of-the-art methods, achieving RMSE values of 14.44 and 13.40 for FD001 and FD003, respectively. Other evaluation criteria, such as MAE and scoring, were also used, with MAE achieving values of 10.49 and 10.68 for FD001 and FD003, respectively. The scoring achieved values of 282.38 and 264.47 for the same sub-datasets. These results highlight the model's promise for improving prognostics and health management (PHM) systems, offering a dependable tool for predictive maintenance in aerospace and related fields. They also demonstrate the effectiveness and superiority of the model in enhancing aviation safety.

PMID:40683914 | DOI:10.1038/s41598-025-09155-z

Acad Radiol. 2025 Jul 19:S1076-6332(25)00636-1. doi: 10.1016/j.acra.2025.06.056. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: To develop and validate a deep learning model based on arterial phase-enhanced CT for predicting the pathological grading of clear cell renal cell carcinoma (ccRCC).

MATERIALS AND METHODS: Data from 564 patients diagnosed with ccRCC from five distinct hospitals were retrospectively analyzed. Patients from centers 1 and 2 were randomly divided into a training set (n=283) and an internal test set (n=122). Patients from centers 3, 4, and 5 served as external validation sets 1 (n=60), 2 (n=38), and 3 (n=61), respectively. A 2D model, a 2.5D model (three-slice input), and a radiomics-based multi-layer perceptron (MLP) model were developed. Model performance was evaluated using the area under the curve (AUC), accuracy, and sensitivity.

RESULTS: The 2.5D model outperformed the 2D and MLP models. Its AUCs were 0.959 (95% CI: 0.9438-0.9738) for the training set, 0.879 (95% CI: 0.8401-0.9180) for the internal test set, and 0.870 (95% CI: 0.8076-0.9334), 0.862 (95% CI: 0.7581-0.9658), and 0.849 (95% CI: 0.7766-0.9216) for the three external validation sets, respectively. The corresponding accuracy values were 0.895, 0.836, 0.827, 0.825, and 0.839. Compared to the MLP model, the 2.5D model achieved significantly higher AUCs (increases of 0.150 [p<0.05], 0.112 [p<0.05], and 0.088 [p<0.05]) and accuracies (increases of 0.077 [p<0.05], 0.075 [p<0.05], and 0.101 [p<0.05]) in the external validation sets.

CONCLUSION: The 2.5D model based on 2.5D CT image input demonstrated improved predictive performance for the WHO/ISUP grading of ccRCC.

PMID:40683765 | DOI:10.1016/j.acra.2025.06.056

J Gastroenterol Hepatol. 2025 Jul 19. doi: 10.1111/jgh.70004. Online ahead of print.

ABSTRACT

BACKGROUND AND AIM: We aimed to develop a deep convolutional neural network (DCNN) that integrates features from multiple sites of the stomach to classify Hp infection status, distinguishing between uninfected, previously infected, and currently infected.

METHODS: Ten deep learning architectures were employed to develop DCNN models using a training dataset comprising 3380 white-light images collected from 676 subjects across eight centers. External validation was conducted with a separate dataset consisting of images from 126 individuals. External testing was subsequently performed to assess and compare the diagnostic efficacy between single-site and multisite fusion DCNN models.

RESULTS: Among these models, the DCNN model using Wide-ResNet emerged as the top performer, achieving a high accuracy of 68.11% (95% confidence interval [CI]: 63.36%-73.09%) with an area under the curve (AUC) of 75.06% (95% CI: 70.22%-80.24%) for noninfection, 69.18% (95% CI: 64.51%-74.03%) for past infection, and 77.04% (95% CI: 72.12%-82.39%) for current infection using images from a single site on the lesser gastric curvature. In comparison, the voting-based multisite fusion DCNN model demonstrated superior accuracy (73.83%, 95% CI: 69.12%-78.65%) and AUC (77.51%, 95% CI: 72.89%-82.59%), particularly notable for noninfection and current infection. Additionally, the DCNN model exhibited heightened sensitivity, specificity, and precision compared to experienced endoscopists.

CONCLUSIONS: The DCNN model, crafted through a voting-based multisite fusion, displayed stellar performance, excelling in the classification of Hp infection status into uninfected and currently infected.

PMID:40682425 | DOI:10.1111/jgh.70004

Angiology. 2025 Jul 19:33197251350182. doi: 10.1177/00033197251350182. Online ahead of print.

ABSTRACT

This study aimed to identify patient characteristics linked to mistaken treatments and major adverse cardiovascular events (MACE) in percutaneous coronary intervention (PCI) for coronary artery disease (CAD) using deep learning-based fractional flow reserve (DEEPVESSEL-FFR, DVFFR). A retrospective cohort of 3,840 PCI patients was analyzed using latent class analysis (LCA) based on eight factors. Mistaken treatment was defined as negative DVFFR patients undergoing revascularization or positive DVFFR patients not receiving it. MACE included all-cause mortality, rehospitalization for unstable angina, and non-fatal myocardial infarction. Patients were classified into comorbidities (Class 1), smoking-drinking (Class 2), and relatively healthy (Class 3) groups. Mistaken treatment was highest in Class 2 (15.4% vs. 6.7%, P < .001), while MACE was highest in Class 1 (7.0% vs. 4.8%, P < .001). Adjusted analyses showed increased mistaken treatment risk in Class 1 (OR 1.96; 95% CI 1.49-2.57) and Class 2 (OR 1.69; 95% CI 1.28-2.25) compared with Class 3. Class 1 also had higher MACE risk (HR 1.53; 95% CI 1.10-2.12). In conclusion, comorbidities and smoking-drinking classes had higher mistaken treatment and MACE risks compared with the relatively healthy class.

PMID:40682405 | DOI:10.1177/00033197251350182

J Magn Reson Imaging. 2025 Jul 19. doi: 10.1002/jmri.70048. Online ahead of print.

ABSTRACT

Hepatocellular carcinoma (HCC) is the most common subtype of primary liver cancer, with significant variability in patient outcomes even within the same stage according to the Barcelona Clinic Liver Cancer staging system. Accurately predicting patient prognosis and potential treatment response prior to therapy initiation is crucial for personalized clinical decision-making. This review focuses on the application of artificial intelligence (AI) in magnetic resonance imaging for guiding individualized treatment strategies in HCC management. Specifically, we emphasize AI-based tools for pre-treatment prediction of therapeutic response and prognosis. AI techniques such as radiomics and deep learning have shown strong potential in extracting high-dimensional imaging features to characterize tumors and liver parenchyma, predict treatment outcomes, and support prognostic stratification. These advances contribute to more individualized and precise treatment planning. However, challenges remain in model generalizability, interpretability, and clinical integration, highlighting the need for standardized imaging datasets and multi-omics fusion to fully realize the potential of AI in personalized HCC care. Evidence level: 5. Technical efficacy: 4.

PMID:40682357 | DOI:10.1002/jmri.70048

Clin Oral Implants Res. 2025 Jul 18. doi: 10.1111/clr.70003. Online ahead of print.

ABSTRACT

OBJECTIVES: To assess the accuracy and time efficiency of manual versus artificial intelligence (AI)-driven tooth segmentation on cone-beam computed tomography (CBCT) images, using AI tools integrated within implant planning software, and to evaluate the impact of artifacts, dental arch, tooth type, and region.

MATERIALS AND METHODS: Fourteen patients who underwent CBCT scans were randomly selected for this study. Using the acquired datasets, 67 extracted teeth were segmented using one manual and two AI-driven tools. The segmentation time for each method was recorded. The extracted teeth were scanned with an intraoral scanner to serve as the reference. The virtual models generated by each segmentation method were superimposed with the surface scan models to calculate volumetric discrepancies.

RESULTS: The discrepancy between the evaluated AI-driven and manual segmentation methods ranged from 0.10 to 0.98 mm, with a mean RMS of 0.27 (0.11) mm. Manual segmentation resulted in less RMS deviation compared to both AI-driven methods (CDX; BSB) (p < 0.05). Significant differences were observed between all investigated segmentation methods, both for the overall tooth area and each region, with the apical portion of the root showing the lowest accuracy (p < 0.05). Tooth type did not have a significant effect on segmentation (p > 0.05). Both AI-driven segmentation methods reduced segmentation time compared to manual segmentation (p < 0.05).

CONCLUSIONS: AI-driven segmentation can generate reliable virtual 3D tooth models, with accuracy comparable to that of manual segmentation performed by experienced clinicians, while also significantly improving time efficiency. To further enhance accuracy in cases involving restoration artifacts, continued development and optimization of AI-driven tooth segmentation models are necessary.

PMID:40682303 | DOI:10.1111/clr.70003

Cardiovasc Diabetol. 2025 Jul 18;24(1):294. doi: 10.1186/s12933-025-02829-y.

ABSTRACT

BACKGROUND: Epicardial adipose tissue (EAT) is associated with cardiometabolic risk in type 2 diabetes (T2D), but its spatial distribution and structural alterations remain understudied. We aim to develop a shape-aware, AI-based method for automated segmentation and morphogeometric analysis of EAT in T2D.

METHODS: A total of 90 participants (45 with T2D and 45 age-, sex-matched controls) underwent cardiac 3D Dixon MRI, enrolled between 2014 and 2018 as part of the sub-study of the Swedish SCAPIS cohort. We developed EAT-Seg, a multi-modal deep learning model incorporating signed distance maps (SDMs) for shape-aware segmentation. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), the 95% Hausdorff distance (HD95), and the average symmetric surface distance (ASSD). Statistical shape analysis combined with partial least squares discriminant analysis (PLS-DA) was applied to point cloud representations of EAT to capture latent spatial variations between groups. Morphogeometric features, including volume, 3D local thickness map, elongation and fragmentation index, were extracted and correlated with PLS-DA latent variables using Pearson correlation. Features with high-correlation were identified as key differentiators and evaluated using a Random Forest classifier.

RESULTS: EAT-Seg achieved a DSC of 0.881, a HD95 of 3.213 mm, and an ASSD of 0.602 mm. Statistical shape analysis revealed spatial distribution differences in EAT between T2D and control groups. Morphogeometric feature analysis identified volume and thickness gradient-related features as key discriminators (r > 0.8, P < 0.05). Random Forest classification achieved an AUC of 0.703.

CONCLUSIONS: This AI-based framework enables accurate segmentation for structurally complex EAT and reveals key morphogeometric differences associated with T2D, supporting its potential as a biomarker for cardiometabolic risk assessment.

PMID:40682091 | DOI:10.1186/s12933-025-02829-y

Respir Res. 2025 Jul 18;26(1):251. doi: 10.1186/s12931-025-03325-3.

ABSTRACT

BACKGROUND: The assessment of diaphragm function is crucial for effective clinical management and the prevention of complications associated with diaphragmatic dysfunction. However, current measurement methodologies rely on manual techniques that are susceptible to human error: How does the performance of an automatic diaphragm measurement system based on a segmentation neural network focusing on diaphragm thickness and excursion compare with existing methodologies?

METHODS: The proposed system integrates segmentation and parameter measurement, leveraging a newly established ultrasound diaphragm dataset. This dataset comprises B-mode ultrasound images and videos for diaphragm thickness assessment, as well as M-mode images and videos for movement measurement. We introduce a novel deep learning-based segmentation network, the Multi-ratio Dilated U-Net (MDRU-Net), to enable accurate diaphragm measurements. The system additionally incorporates a comprehensive implementation plan for automated measurement.

RESULTS: Automatic measurement results are compared against manual assessments conducted by clinicians, revealing an average error of 8.12% in diaphragm thickening fraction measurements and a mere 4.3% average relative error in diaphragm excursion measurements. The results indicate overall minor discrepancies and enhanced potential for clinical detection of diaphragmatic conditions. Additionally, we design a user-friendly automatic measurement system for assessing diaphragm parameters and an accompanying method for measuring ultrasound-derived diaphragm parameters.

CONCLUSIONS: In this paper, we constructed a diaphragm ultrasound dataset of thickness and excursion. Based on the U-Net architecture, we developed an automatic diaphragm segmentation algorithm and designed an automatic parameter measurement scheme. A comparative error analysis was conducted against manual measurements. Overall, the proposed diaphragm ultrasound segmentation algorithm demonstrated high segmentation performance and efficiency. The automatic measurement scheme based on this algorithm exhibited high accuracy, eliminating subjective influence and enhancing the automation of diaphragm ultrasound parameter assessment, thereby providing new possibilities for diaphragm evaluation.

PMID:40682068 | DOI:10.1186/s12931-025-03325-3

BMC Med Inform Decis Mak. 2025 Jul 18;25(1):270. doi: 10.1186/s12911-025-03093-6.

ABSTRACT

BACKGROUND: Predicting associations between microbes and diseases is crucial for clinical diagnosis and therapy. However, biological experiments are time-intensive, necessitating efficient computational models. Traditional models rely on existing microbe-disease associations, but limited data often restricts their effectiveness. This scarcity of information hinders the construction of a comprehensive association network, prompting the need for innovative solutions.

METHODS: We propose RKGATMDA, a deep learning framework for microbe-disease association prediction. Utilizing a graph attention network, RKGATMDA learns representations from the microbe-disease association network. To address the limitation of insufficient association information, we introduce Random K-Nearest Neighbors to uncover latent relationships, enhancing representation learning. During each training iteration, associations are expanded based on attention scores, and a multi-head attention mechanism integrates diverse features, enabling RKGATMDA to capture comprehensive interactions between microbes and diseases.

RESULTS: Results Experimental results show that RKGATMDA achieves AUC values of 0.8906 in 5-fold cross-validation, 0.8999 in global leave-one-out cross-validation, and 0.7246 in local leave-one-out cross-validation, outperforming previous methods such as ABHMDA, KATZHMDA, LRLSHMDA, BiRWHMDA, and NTSHMDA. Case studies on asthma, colon cancer, and colorectal carcinoma further validate its predictive power.

CONCLUSION: Our findings demonstrate that RKGATMDA effectively predicts microbe-disease associations, with at least 9 out of the top 10 prediction pairs validated by biological evidence. This highlights the potential of RKGATMDA as a valuable tool in microbial-disease research, offering a promising approach for identifying novel associations and advancing our understanding of microbial pathogenesis.

PMID:40682015 | DOI:10.1186/s12911-025-03093-6

BMC Bioinformatics. 2025 Jul 18;26(1):183. doi: 10.1186/s12859-025-06230-0.

ABSTRACT

BACKGROUND: Reliable identification of cancer subtypes is crucial for devising personalized treatment strategies. Integrating multi-omics data has proven to be an effective method for analyzing cancer subtypes. By combining molecular information across various layers, a more comprehensive understanding of biological characteristics and disease mechanisms can be achieved.

RESULTS: We propose DEGCN, a novel deep learning model that integrates a three-channel Variational Autoencoder (VAE) for multi-omics dimensionality reduction and a densely connected Graph Convolutional Network (GCN) for effective subtype classification. DEGCN leverages the complementary strengths of non-linear feature extraction and graph-based relational learning, enabling accurate and robust classification of renal cancer subtypes. Experimental results demonstrate that DEGCN achieves a cross-validated classification accuracy of 97.06% ± 2.04% on renal cancer data, outperforming conventional machine learning algorithms and state-of-the-art deep learning models. Moreover, its generalization ability is validated on breast and gastric cancer datasets from TCGA, with cross-validated classification accuracies of 89.82% ± 2.29% and 88.64% ± 5.24%, respectively, indicating strong cross-cancer predictive performance.

CONCLUSION: The study highlights the outstanding performance of DEGCN in heterogeneous data integration and classification accuracy, demonstrating the model's potential in cancer subtype prediction and its application in guiding clinical treatment.

PMID:40681997 | DOI:10.1186/s12859-025-06230-0

NPJ Digit Med. 2025 Jul 18;8(1):456. doi: 10.1038/s41746-025-01848-z.

ABSTRACT

This meta-analysis evaluated diagnostic performance of deep learning (DL) algorithms using whole slide images (WSIs) for detecting microsatellite instability-high (MSI-H) in colorectal cancer (CRC). PubMed, Embase, and Web of Science were searched until January 2025. Nineteen studies comprising 33,383 samples were included. Bivariate random-effects models calculated pooled sensitivity/specificity with 95% CIs. The revised QUADAS-2 tool was used for quality assessment. Pooled patient-based internal validation showed a sensitivity of 0.88 and specificity of 0.86, while external validation revealed higher sensitivity of 0.93 but lower specificity of 0.71. Image-based analysis showed similar accuracy. Meta-regression identified center, reference standard, and tile size as major sources of heterogeneity, with no significant differences observed between internal and external performance. Overall, DL algorithms demonstrate excellent sensitivity in detecting MSI-H; however, their lower specificity in external validation suggests overfitting and highlights the need for algorithm standardization to improve generalizability and clinical utility.

PMID:40681867 | DOI:10.1038/s41746-025-01848-z

Sci China Life Sci. 2025 Jul 16. doi: 10.1007/s11427-025-3002-y. Online ahead of print.

NO ABSTRACT

PMID:40681822 | DOI:10.1007/s11427-025-3002-y

Water Res. 2025 Jul 10;286:123928. doi: 10.1016/j.watres.2025.123928. Online ahead of print.

ABSTRACT

Environmental and water quality monitoring increasingly relies on high-frequency data streams from sensor networks, yet anomalies in these datasets can compromise their reliability. Here we introduce two novel unsupervised methods for anomaly detection in spatio-temporal sensor arrays specifically designed for highly structured datasets such as those obtained by networks of sensors in rivers. The first is a dynamic Bayesian spatio-temporal model using a reduced rank Gaussian process, and the second is a deep learning architecture called Spatio-Temporal Attention-based LSTM for River Networks. We rigorously evaluate both methods through comprehensive simulation benchmarks incorporating diverse anomaly types common in environmental data. Our comparative analysis reveals the strengths and limitations of each approach, demonstrating superior performance over existing methods in both accuracy and computational efficiency. We further introduce an ensemble method that synergistically combines the strengths of both approaches. Our framework addresses the growing need for robust, efficient algorithms and computational methods for monitoring complex ecosystems, advancing spatio-temporal anomaly detection in environmental applications. By providing detailed implementation guidelines and open-source code, we enable immediate application by ecological and environmental scientists and practitioners, facilitating improved monitoring and enhanced decision-making in river network management.

PMID:40680322 | DOI:10.1016/j.watres.2025.123928

Hepatology. 2025 Aug 1;82(2):275-276. doi: 10.1097/HEP.0000000000001410. Epub 2025 Jul 18.

NO ABSTRACT

PMID:40680277 | DOI:10.1097/HEP.0000000000001410

Chem Res Toxicol. 2025 Jul 18. doi: 10.1021/acs.chemrestox.5c00152. Online ahead of print.

ABSTRACT

Under REACH, mutagenicity assessment relies on in vitro testing (gene mutation test in bacteria and/or mammalian cells, as well as chromosomal aberration or micronucleus assays in mammalian cells) followed by in vivo testing if necessary. This study explored the possibility of using the inherent correlation between these in vitro assays to create multi-task deep learning models and examine if they outperform single-task models. An extensive genotoxicity dataset with over 12,000 substances was compiled, including algorithmically curated REACH data and information from several public sources. Genotoxicity information was also retrieved from ToxValDB and literature sources to construct external (hold-out) test sets for a stringent assessment of the models' generalized performance. A range of single-task and multi-task models were investigated from classical machine learning techniques and chemical fingerprints to deep learning methods using graphs for molecular structure representation. The best deep learning single-task model achieved a cross-validation balanced accuracy of 73-84% for the four in vitro assays and exceeded classical machine learning by 2-8%. Gene mutation detection for specific bacterial strains and metabolic activation modes exhibited balanced accuracy 82-85%, with improvements ranging from 7% to 12%. Multi-task deep learning models for specific bacterial strains and metabolic activation modes had on average 8% higher cross-validation test balanced accuracy than single-task models but were comparable when assay outcomes were aggregated. The best deep learning models for specific bacterial strains and metabolic activation modes showed external balanced accuracy of 72-78 % when there were at least 200 positives and 200 negatives. The dimensionality-reduced molecular embeddings from graph neural network models were able to distinguish positives from negatives and cluster structures that trigger known genotoxicity structural alerts. The models were also used to identify structural moieties linked to predicted negative genotoxicity in bacteria and positive genotoxicity in mammalian cells.

PMID:40680271 | DOI:10.1021/acs.chemrestox.5c00152

J Med Internet Res. 2025 Jul 18;27:e73528. doi: 10.2196/73528.

ABSTRACT

BACKGROUND: With the widespread application of machine learning (ML) in the diagnosis and treatment of colorectal cancer (CRC), some studies have investigated the use of ML techniques for the diagnosis of KRAS (Kirsten rat sarcoma) mutation. Nevertheless, there is scarce evidence from evidence-based medicine to substantiate its efficacy.

OBJECTIVE: Our study was carried out to systematically review the performance of ML models developed using different modeling approaches, in diagnosing KRAS mutations in CRC. We aim to offer evidence-based foundations for the development and enhancement of future intelligent diagnostic tools.

METHODS: PubMed, Cochrane Library, Embase, and Web of Science were systematically retrieved, with the search cutoff date set to December 22, 2024. The encompassed studies are publicly published research papers that use ML to diagnose KRAS gene mutations in CRC. The risk of bias in the encompassed models was evaluated via the PROBAST (Prediction Model Risk of Bias Assessment Tool). A meta-analysis of the model's concordance index (c-index) was performed, and a bivariate mixed-effects model was used to summarize sensitivity and specificity based on diagnostic contingency tables.

RESULTS: A total of 43 studies involving 10,888 patients were included. The modeling variables were derived from clinical characteristics, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography/computed tomography, and pathological histology. In the validation cohort, for the ML model developed based on CT radiomic features, the c-index, sensitivity, and specificity were 0.87 (95% CI 0.84-0.90), 0.85 (95% CI 0.80-0.89), and 0.83 (95% CI 0.73-0.89), respectively. For the model developed using MRI radiomic features, the c-index, sensitivity, and specificity were 0.77 (95% CI 0.71-0.83), 0.78 (95% CI 0.72-0.83), and 0.73 (95% CI 0.63-0.81), respectively. For the ML model developed based on positron emission tomography/computed tomography radiomic features, the c-index, sensitivity, and specificity were 0.84 (95% CI 0.77-0.90), 0.73, and 0.83, respectively. Notably, the deep learning (DL) model based on pathological images demonstrated a c-index, sensitivity, and specificity of 0.96 (95% CI 0.94-0.98), 0.83 (95% CI 0.72-0.91), and 0.87 (95% CI 0.77-0.92), respectively. The DL model MRI-based model showed a c-index of 0.93 (95% CI 0.90-0.96), sensitivity of 0.85 (95% CI 0.75-0.91), and specificity of 0.83 (95% CI 0.77-0.88).

CONCLUSIONS: ML is highly accurate in diagnosing KRAS mutations in CRC, and DL models based on MRI and pathological images exhibit particularly strong diagnosis accuracy. More broadly applicable DL-based diagnostic tools may be developed in the future. However, the clinical application of DL models remains relatively limited at present. Therefore, future research should focus on increasing sample sizes, improving model architectures, and developing more advanced DL models to facilitate the creation of highly efficient intelligent diagnostic tools for KRAS mutation diagnosis in CRC.

PMID:40680189 | DOI:10.2196/73528

Sci Adv. 2025 Jul 18;11(29):eadv6484. doi: 10.1126/sciadv.adv6484. Epub 2025 Jul 18.

ABSTRACT

The factors controlling the spatial distribution and temporal evolution of earthquake swarms in volcanic systems remain unclear. We leverage leading-edge deep learning algorithms and a detailed three-dimensional velocity model to construct a 15-year high-resolution earthquake catalog of the Yellowstone caldera region. More than half of the region's earthquakes are clustered into swarm-like families characterized by episodes of hypocenter expansion and migration. Adjacent earthquake swarms, separated by long quiescent periods, are found to be a dominant feature. We suggest that these swarms are controlled by the interplay between slowly diffusing aqueous fluids and rapid episodic fluid injections, which may result from the breaking of permeability seals. Our analyses also indicate that clustered seismicity beneath the caldera occurs on relatively immature, rougher fault structures, compared to more planar faults outside. Our results provide additional context for understanding seismicity in hydrothermal systems, highlighting the key role played by long-term fluid diffusion processes in driving the occurrence of earthquake swarms.

PMID:40680135 | DOI:10.1126/sciadv.adv6484