Zhonghua Bing Li Xue Za Zhi. 2025 Jan 8;54(1):59-65. doi: 10.3760/cma.j.cn112151-20240712-00455.

ABSTRACT

Objective: To investigate the prognostic value of deep learning-based automated quantification of tumor-stroma ratio (TSR) in patients undergoing neoadjuvant therapy (NAT) for breast cancer. Methods: Specimens were collected from 209 breast cancer patients who received NAT at Renmin Hospital of Wuhan University from October 2019 to June 2023. TSR levels in pre-NAT biopsy specimens were automatically computed using a deep learning algorithm and categorized into low stroma (TSR≤30%), intermediate stroma (TSR 30% to ≤60%), and high stroma (TSR>60%) groups. Residual cancer burden (RCB) grading of post-NAT surgical specimens was determined to compare the relationship between TSR expression levels and RCB grades. The correlation of TSR with NAT efficacy was analyzed, and the association between TSR expression and patient prognosis was further investigated. Results: There were 85 cases with low stroma (TSR≤30%), 93 cases with intermediate stroma (TSR 30% to ≤60%), and 31 cases with high stroma (TSR>60%). Different TSR expression levels showed significant differences between various RCB grades (P<0.05). Logistic univariate and multivariate analyses showed that TSR was a risk factor for obtaining a complete pathological remission from neoadjuvant therapy for breast cancer when it was used as a continuous variable (P<0.05); COX regression and survival analyses showed that the lower the percentage of tumorigenic mesenchyme was, the better the prognosis of the patient was (P<0.05). Conclusions: The deep learning-based model enables automatic and accurate quantification of TSR. A lower pre-treatment tumoral stroma is associated with a lower RCB score and a higher rate of pathologic complete response, indicating that TSR can predict the efficacy of neoadjuvant therapy in breast cancer and thus holds prognostic significance. Therefore, TSR may serve as a biomarker for predicting therapeutic outcomes in breast cancer neoadjuvant therapy.

PMID:39762173 | DOI:10.3760/cma.j.cn112151-20240712-00455

Inj Prev. 2025 Jan 6:ip-2024-045314. doi: 10.1136/ip-2024-045314. Online ahead of print.

ABSTRACT

BACKGROUND: Improving the infrastructure for drug overdose surveillance is critical for identifying new threats and responding to emerging trends. We aimed to develop a prototype tool using the principles of natural language processing that can extract information from the death records of drug overdose victims.

METHODS: Data were obtained from the Violent Death Reporting System on drug overdose deaths. Narratives were manually labelled for 12 attributes of interest, totalling 82 labels about the circumstances of the overdose. Narratives were passed through the 'Excel Extractor' to identify and extract a target phrase and subsequently map the extracted phrase to predetermined code values. The output from the Excel Extractor was compared with manually labelled data to determine accuracy. Performance was compared against multiple machine learning models.

RESULTS: The Excel Extractor performed well across the attributes of interest, achieving an F1 Score over 0.8 on nine of the 12 attributes. The Excel Extractor was the highest performing model on seven of the 12 attributes. The Excel Extractor achieved an F1 Score of 0.8 or higher on 46 of 82 (56%) of the labels, and a score of 0.9 or higher on nearly one-third (25 out of 82) of the labels.

CONCLUSION: This work demonstrates it is feasible to develop a spreadsheet-formula-based natural language processing tool to accurately extract information about drug overdose deaths from narratives; for most attributes, a rule-based search performs well or better than deep learning. The Excel Extractor has the potential to streamline data abstraction for epidemiologists gathering data about drug overdose deaths.

PMID:39762007 | DOI:10.1136/ip-2024-045314

Phys Med Biol. 2025 Jan 6. doi: 10.1088/1361-6560/ada67f. Online ahead of print.

ABSTRACT

Radiation dose and diagnostic image quality are opposing constraints in x-ray CT. Conventional methods do not fully account for organ-level radiation dose and noise when considering radiation risk and clinical task. In this work, we develop a pipeline to generate individualized organ-specific dose and noise at desired dose levels from clinical CT scans. &#xD;Approach: To estimate organ-specific dose and noise, we compute dose maps, noise maps at desired dose levels and organ segmentations. In our pipeline, dose maps are generated using Monte Carlo simulation. The noise map is obtained by scaling the inserted noise in synthetic low-dose emulation in order to avoid anatomical structures, where the scaling coefficients are empirically calibrated. Organ segmentations are generated by a deep learning-based method (TotalSegmentator). The proposed noise model is evaluated on a clinical dataset of 12 CT scans, a phantom dataset of 3 uniform phantom scans, and a cross-site dataset of 26 scans. The accuracy of deep learning-based segmentations for organ-level dose and noise estimates was tested using a dataset of 41 cases with expert segmentations of six organs: lungs, liver, kidneys, bladder, spleen, and pancreas.&#xD;Main Results: The empirical noise model performs well, with an average RMSE approximately 1.5 HU and an average relative RMSE approximately 5% across different dose levels. The segmentation from TotalSegmentator yielded a mean Dice score of 0.8597 across the six organs (max=0.9315 in liver, min=0.6855 in pancreas). The resulting error in organ-level dose and noise estimation was less than 2% for most organs.&#xD;Significance: The proposed pipeline can output individualized organ-specific dose and noise estimates accurately for personalized protocol evaluation and optimization. It is fully automated and can be scalable to large clinical datasets. This pipeline can be used to optimize image quality for specific organs and thus clinical tasks, without adversely affecting overall radiation dose.

PMID:39761638 | DOI:10.1088/1361-6560/ada67f

BMC Med Inform Decis Mak. 2025 Jan 6;25(1):6. doi: 10.1186/s12911-024-02845-0.

ABSTRACT

BACKGROUND: The diagnosis and treatment of epilepsy continue to face numerous challenges, highlighting the urgent need for the development of rapid, accurate, and non-invasive methods for seizure detection. In recent years, advancements in the analysis of electroencephalogram (EEG) signals have garnered widespread attention, particularly in the area of seizure recognition.

METHODS: A novel hybrid deep learning approach that combines feature fusion for efficient seizure detection is proposed in this study. First, the Discrete Wavelet Transform (DWT) is applied to perform a five-level decomposition of the raw EEG signals, from which time-frequency and nonlinear features are extracted from the decomposed sub-bands. To eliminate redundant features, Support Vector Machine-Recursive Feature Elimination (SVM-RFE) is employed to select the most distinctive features for fusion. Finally, seizure states are classified using Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-Bi-LSTM).

RESULTS: The method was rigorously validated on the Bonn and New Delhi datasets. In the binary classification tasks, both the D-E group (Bonn dataset) and the Interictal-Ictal group (New Delhi dataset) achieved 100% accuracy, 100% sensitivity, 100% specificity, 100% precision, and 100% F1-score. In the three-class classification task A-D-E on the Bonn dataset, the model performed excellently, achieving 96.19% accuracy, 95.08% sensitivity, 97.34% specificity, 97.49% precision, and 96.18% F1-score. In addition, the proposed method was further validated on the larger and more clinically relevant CHB-MIT dataset, achieving average metrics of 98.43% accuracy, 97.84% sensitivity, 99.21% specificity, 99.14% precision, and an F1 score of 98.39%. Compared to existing literature, our method outperformed several recent studies in similar classification tasks, underscoring the effectiveness and advancement of the approach presented in this research.

CONCLUSION: The findings indicate that the proposed method demonstrates a high level of effectiveness in detecting seizures, which is a crucial aspect of managing epilepsy. By improving the accuracy of seizure detection, this method has the potential to significantly enhance the process of diagnosing and treating individuals affected by epilepsy. This advancement could lead to more tailored treatment plans, timely interventions, and ultimately, better quality of life for patients.

PMID:39762881 | DOI:10.1186/s12911-024-02845-0

J Transl Med. 2025 Jan 6;23(1):13. doi: 10.1186/s12967-024-06034-5.

ABSTRACT

BACKGROUND: First-line treatment for advanced gastric adenocarcinoma (GAC) with human epidermal growth factor receptor 2 (HER2) is trastuzumab combined with chemotherapy. In clinical practice, HER2 positivity is identified through immunohistochemistry (IHC) or fluorescence in situ hybridization (FISH), whereas deep learning (DL) can predict HER2 status based on tumor histopathological features. However, it remains uncertain whether these deep learning-derived features can predict the efficacy of anti-HER2 therapy.

METHODS: We analyzed a cohort of 300 consecutive surgical specimens and 101 biopsy specimens, all undergoing HER2 testing, along with 41 biopsy specimens receiving trastuzumab-based therapy for HER2-positive GAC.

RESULTS: We developed a convolutional neural network (CNN) model using surgical specimens that achieved an area under the curve (AUC) value of 0.847 in predicting HER2 amplification, and achieved an AUC of 0.903 in predicting HER2 status specifically in patients with HER2 2 + expression. The model also predicted HER2 status in gastric biopsy specimens, achieving an AUC of 0.723. Furthermore, our classifier was trained using 41 HER2-positive gastric biopsy specimens that had undergone trastuzumab treatment, our model demonstrated an AUC of 0.833 for the (CR + PR) / (SD + PD) subgroup.

CONCLUSION: This work explores an algorithm that utilizes hematoxylin and eosin (H&E) staining to accurately predict HER2 status and assess the response to trastuzumab in GAC, potentially facilitating clinical decision-making.

PMID:39762854 | DOI:10.1186/s12967-024-06034-5

BMC Med Imaging. 2025 Jan 6;25(1):6. doi: 10.1186/s12880-024-01522-y.

ABSTRACT

Invasive breast cancer diagnosis and treatment planning require an accurate assessment of human epidermal growth factor receptor 2 (HER2) expression levels. While immunohistochemical techniques (IHC) are the gold standard for HER2 evaluation, their implementation can be resource-intensive and costly. To reduce these obstacles and expedite the procedure, we present an efficient deep-learning model that generates high-quality IHC-stained images directly from Hematoxylin and Eosin (H&E) stained images. We propose a new IHC-GAN that enhances the Pix2PixHD model into a dual generator module, improving its performance and simplifying its structure. Furthermore, to strengthen feature extraction for HE-stained image classification, we integrate MobileNetV3 as the backbone network. The extracted features are then merged with those generated by the generator to improve overall performance. Moreover, the decoder's performance is enhanced by providing the related features from the classified labels by incorporating the adaptive instance normalization technique. The proposed IHC-GAN was trained and validated on a comprehensive dataset comprising 4,870 registered image pairs, encompassing a spectrum of HER2 expression levels. Our findings demonstrate promising results in translating H&E images to IHC-equivalent representations, offering a potential solution to reduce the costs associated with traditional HER2 assessment methods. We extensively validate our model and the current dataset. We compare it with state-of-the-art techniques, achieving high performance using different evaluation metrics, showing 0.0927 FID, 22.87 PSNR, and 0.3735 SSIM. The proposed approach exhibits significant enhancements over current GAN models, including an 88% reduction in Frechet Inception Distance (FID), a 4% enhancement in Learned Perceptual Image Patch Similarity (LPIPS), a 10% increase in Peak Signal-to-Noise Ratio (PSNR), and a 45% reduction in Mean Squared Error (MSE). This advancement holds significant potential for enhancing efficiency, reducing manpower requirements, and facilitating timely treatment decisions in breast cancer care.

PMID:39762786 | DOI:10.1186/s12880-024-01522-y

J Imaging Inform Med. 2025 Jan 6. doi: 10.1007/s10278-024-01372-8. Online ahead of print.

ABSTRACT

Although the relationships between basic clinical parameters and white matter hyperintensity (WMH) have been studied, the associations between vascular factors and WMH volume in general populations remain unclear. We investigated the associations between clinical parameters including comprehensive vascular factors and WMH in two large general populations. This retrospective, cross-sectional study involved two populations: individuals who underwent general health examinations at the Asan Medical Center (AMC) and participants from a regional cohort, the Korean Genome and Epidemiology Study (KoGES). WMH volume was quantified using the deep learning model nnU-Net. The associations between vascular factors and WMH volume were analyzed using multivariate linear regression. Individuals in the AMC cohort (n = 7471) had a mean [SD] age of 58.0 [9.2] years, and the KoGES participants (n = 2511), 59.2 [6.8] years. The normalized and logit-transformed WMH volumes for the AMC and KoGES were - 8.5 [1.3] and - 7.9 [1.2], respectively. The presence of carotid plaque, brachial-ankle pulse wave velocity, Agaston score, and coronary artery stenosis were associated with WMH volume after adjustments (AMC: carotid plaque β 0.13; 95% CI, 0.06-0.20; p < 0.001, baPWV β 0.001; CI 0-0.001; p < 0.001, Agaston score β 0.0003; CI 0.0001-0.0005; p < 0.001, minimal-to-mild coronary artery stenosis β 0.20; CI 0.12-0.29; p < 0.001, moderate-to-severe coronary artery stenosis β 0.30; CI 0.15-0.44; p < 0.001, KoGES: carotid plaque β 0.15; CI 0.02-0.27; p = 0.02, baPWV β 0.0004; CI 0-0.001; p = 0.001). Vascular parameters, reflecting atherosclerotic changes in carotid and coronary arteries and arterial stiffness, were independently associated with WMH volume in the general population.

PMID:39762547 | DOI:10.1007/s10278-024-01372-8

J Imaging Inform Med. 2025 Jan 6. doi: 10.1007/s10278-024-01379-1. Online ahead of print.

ABSTRACT

Nailfold microcirculation examination is crucial for the early differential diagnosis of diseases and indicating their severity. In particular, panoramic nailfold flow velocity measurements can provide direct quantitative indicators for the study of vascular diseases and technical support to assess vascular health. Previously, nailfold imaging equipment was limited by a small field of view. Therefore, research on nailfold flow velocity measurement primarily focused on improving the accuracy of single-vessel flow velocity results, while there were few studies on nailfold panoramic flow velocity. Furthermore, with improvements in the imaging field of view and the increasing clinical demand for speed in obtaining nailfold parameter results, doctors do not have time to crop videos to obtain flow velocity results. Therefore, research on nailfold panoramic flow velocity measurement is crucial. This study presents a panoramic nailfold flow velocity measurement method based on enhanced plasma gap information. In contrast to previous methods, the use of a deep learning model to decompose the panoramic flow velocity measurement task into several vessel flow velocity measurement tasks is proposed herein. For improved accuracy, a plasma gap information enhancement method is proposed, using the frame difference to enhance the position movement information of plasma gaps in videos. The t-test results show that the Pearson correlation coefficient between the results of the proposed method and those manually calculated by experts is 0.992 (t = - 0.0889, p = 0.929; > 0.05), with an average error of 2.137%. Therefore, there is no significant difference between the results obtained by the proposed method proposed and the manually calculated results. The feasibility experiment demonstrates that the proposed method can concurrently obtain the flow rate results of 13 nailfold blood vessels. Finally, the proposed method provides an efficient solution for panoramic flow velocity measurement of large-field nailfold multi-vessel videos.

PMID:39762546 | DOI:10.1007/s10278-024-01379-1

J Imaging Inform Med. 2025 Jan 6. doi: 10.1007/s10278-024-01377-3. Online ahead of print.

ABSTRACT

A scoping review was conducted to investigate the role of radiological imaging, particularly high-resolution computed tomography (HRCT), and artificial intelligence (AI) in diagnosing and prognosticating idiopathic pulmonary fibrosis (IPF). Relevant studies from the PubMed database were selected based on predefined inclusion and exclusion criteria. Two reviewers assessed study quality and analyzed data, estimating heterogeneity and publication bias. The analysis primarily focused on deep learning approaches for feature extraction from HRCT images, aiming to enhance diagnostic accuracy and efficiency. Radiomics, utilizing quantitative features extracted from images, were computed using various tools to improve precision in analysis. Validation methods such as k-fold cross-validation were employed to assess model robustness and generalizability. Findings revealed that radiologic patterns in interstitial lung disease hold prognostic significance for patient survival. However, the additional prognostic value of quantitative assessment of fibrosis extent remains uncertain. IPF poses a substantial challenge in respiratory medicine, necessitating advanced diagnostic and prognostic tools. Radiomics emerges as a valuable asset, offering insights into disease characteristics and aiding in disease classification. It contributes to understanding underlying pathophysiological processes, facilitating more effective management of pulmonary disorders. Future research should focus on clarifying the additional prognostic value of quantitative assessment and further refining AI-based diagnostic and prognostic models for IPF.

PMID:39762544 | DOI:10.1007/s10278-024-01377-3

Sci Rep. 2025 Jan 6;15(1):1027. doi: 10.1038/s41598-024-84504-y.

ABSTRACT

This study utilizes the Breast Ultrasound Image (BUSI) dataset to present a deep learning technique for breast tumor segmentation based on a modified UNet architecture. To improve segmentation accuracy, the model integrates attention mechanisms, such as the Convolutional Block Attention Module (CBAM) and Non-Local Attention, with advanced encoder architectures, including ResNet, DenseNet, and EfficientNet. These attention mechanisms enable the model to focus more effectively on relevant tumor areas, resulting in significant performance improvements. Models incorporating attention mechanisms outperformed those without, as reflected in superior evaluation metrics. The effects of Dice Loss and Binary Cross-Entropy (BCE) Loss on the model's performance were also analyzed. Dice Loss maximized the overlap between predicted and actual segmentation masks, leading to more precise boundary delineation, while BCE Loss achieved higher recall, improving the detection of tumor areas. Grad-CAM visualizations further demonstrated that attention-based models enhanced interpretability by accurately highlighting tumor areas. The findings denote that combining advanced encoder architectures, attention mechanisms, and the UNet framework can yield more reliable and accurate breast tumor segmentation. Future research will explore the use of multi-modal imaging, real-time deployment for clinical applications, and more advanced attention mechanisms to further improve segmentation performance.

PMID:39762417 | DOI:10.1038/s41598-024-84504-y

Sci Rep. 2025 Jan 6;15(1):999. doi: 10.1038/s41598-024-84049-0.

ABSTRACT

We developed a deep learning-based extraction of electrocardiographic (ECG) waves from ballistocardiographic (BCG) signals and explored their use in R-R interval (RRI) estimation. Preprocessed BCG and reference ECG signals were inputted into the bidirectional long short-term memory network to train the model to minimize the loss function of the mean squared error between the predicted ECG (pECG) and genuine ECG signals. Using a dataset acquired with polyvinylidene fluoride and ECG sensors in different recumbent positions from 18 participants, we generated pECG signals from preprocessed BCG signals using the learned model and evaluated the RRI estimation performance by comparing the predicted RRI with the reference RRI obtained from the ECG signal using a leave-one-subject-out cross-validation scheme. A mean absolute error (MAE) of 0.034 s was achieved for the beat-to-beat interval accuracy. To further test the generalization ability of the learned model trained with a short-term-recorded dataset, we collected long-term overnight recordings of BCG signals from 12 different participants and performed validation. The beat-to-beat interval correlation between BCG and ECG signals was 0.82 ± 0.06 with an average MAE of 0.046 s, showing practical performance for long-term measurement of RRIs. These results suggest that the proposed approach can be used for continuous heart rate monitoring in a home environment.

PMID:39762351 | DOI:10.1038/s41598-024-84049-0

Sci Rep. 2025 Jan 6;15(1):977. doi: 10.1038/s41598-025-85275-w.

ABSTRACT

Bearings are critical in mechanical systems, as their health impacts system reliability. Proactive monitoring and diagnosing of bearing faults can prevent significant safety issues. Among various diagnostic methods that analyze bearing vibration signals, deep learning is notably effective. However, bearings often operate in noisy environments, especially during failures, which poses a challenge to most current deep learning methods that assume noise-free data. Therefore, this paper designs a Multi-Location Multi-Scale Multi-Level Information Attention Activation Network (MLSCA-CW) with excellent performance in different kinds of strong noise environments by combining soft threshold, self-activation, and self-attention mechanisms. The model has enhanced filtering performance and multi-location information fusion ability. Our comparative and ablation experiments demonstrate that the model's components, including the multi-location and multi-scale vibration extraction module, soft threshold noise filtering module, multi-scale self-activation mechanism, and layer attention mechanism, are highly effective in filtering noise from various locations and extracting multi-dimensional features. The MLSCA-CW model achieves 92.02% accuracy against various strong noise disturbance and outperforms SOTA methods under challenging working conditions in CWRU dataset.

PMID:39762349 | DOI:10.1038/s41598-025-85275-w

Sci Data. 2025 Jan 6;12(1):18. doi: 10.1038/s41597-025-04372-7.

ABSTRACT

Arctic permafrost is undergoing rapid changes due to climate warming in high latitudes. Retrogressive thaw slumps (RTS) are one of the most abrupt and impactful thermal-denudation events that change Arctic landscapes and accelerate carbon feedbacks. Their spatial distribution remains poorly characterised due to time-intensive conventional mapping methods. While numerous RTS studies have published standalone digitisation datasets, the lack of a centralised, unified database has limited their utilisation, affecting the scale of RTS studies and the generalisation ability of deep learning models. To address this, we established the Arctic Retrogressive Thaw Slumps (ARTS) dataset containing 23,529 RTS-present and 20,434 RTS-absent digitisations from 20 standalone datasets. We also proposed a Data Curation Framework as a working standard for RTS digitisations. This dataset is designed to be comprehensive, accessible, contributable, and adaptable for various RTS-related studies. This dataset and its accompanying curation framework establish a foundation for enhanced collaboration in RTS research, facilitating standardised data sharing and comprehensive analyses across the Arctic permafrost research community.

PMID:39762331 | DOI:10.1038/s41597-025-04372-7

J Chem Inf Model. 2025 Jan 6. doi: 10.1021/acs.jcim.4c01365. Online ahead of print.

ABSTRACT

Predicting protein-protein interactions (PPIs) is crucial for advancing drug discovery. Despite the proposal of numerous advanced computational methods, these approaches often suffer from poor usability for biologists and lack generalization. In this study, we designed a deep learning model based on a coattention mechanism that was capable of both PPI and site prediction and used this model as the foundation for PPI-CoAttNet, a user-friendly, multifunctional web server for PPI prediction. This platform provides comprehensive services for online PPI model training, PPI and site prediction, and prediction of interactions with proteins associated with highly prevalent cancers. In our Homo sapiens test set for PPI prediction, PPI-CoAttNet achieved an AUC of 0.9841 and an F1 score of 0.9440, outperforming most state-of-the-art models. Additionally, these results are generated in real time, delivering outcomes within minutes. We also evaluated PPI-CoAttNet for downstream tasks, including novel E3 ligase scoring, demonstrating outstanding accuracy. We believe that this tool will empower researchers, especially those without computational expertise, to leverage AI for accelerating drug development.

PMID:39761551 | DOI:10.1021/acs.jcim.4c01365

J Comput Assist Tomogr. 2024 Nov 13. doi: 10.1097/RCT.0000000000001693. Online ahead of print.

ABSTRACT

OBJECTIVES: The aims of the study are to predict lung function impairment in patients with connective tissue disease (CTD)-associated interstitial lung disease (ILD) through computed tomography (CT) quantitative analysis parameters based on CT deep learning model and density threshold method and to assess the severity of the disease in patients with CTD-ILD.

METHODS: We retrospectively collected chest high-resolution CT images and pulmonary function test results from 105 patients with CTD-ILD between January 2021 and December 2023 (patients staged according to the gender-age-physiology [GAP] system), including 46 males and 59 females, with a median age of 64 years. Additionally, we selected 80 healthy controls (HCs) with matched sex and age, who showed no abnormalities in their chest high-resolution CT. Based on our previously developed RDNet analysis model, the proportion of the lung occupied by reticulation, honeycombing, and total interstitial abnormalities in CTD-ILD patients (ILD% = total interstitial abnormal volume/total lung volume) were calculated. Using the Pulmo-3D software with a threshold segmentation method of -260 to -600, the overall interstitial abnormal proportion (AA%) and mean lung density were obtained. The correlations between CT quantitative analysis parameters and pulmonary function indices were evaluated using Spearman or Pearson correlation coefficients. Stepwise multiple linear regression analysis was used to identify the best CT quantitative predictors for different pulmonary function parameters. Independent risk factors for GAP staging were determined using multifactorial logistic regression. The area under the ROC curve (AUC) differentiated between the CTD-ILD groups and HCs, as well as among GAP stages. The Kruskal-Wallis test was used to compare the differences in pulmonary function indices and CT quantitative analysis parameters among CTD-ILD groups.

RESULTS: Among 105 CTD-ILD patients (58 in GAP I, 36 in GAP II, and 11 in GAP III), results indicated that AA% distinguished between CTD-ILD patients and HCs with the highest AUC value of 0.974 (95% confidence interval: 0.955-0.993). With a threshold set at 9.7%, a sensitivity of 98.7% and a specificity of 89.5% were observed. Both honeycombing and ILD% showed statistically significant correlations with pulmonary function parameters, with honeycombing displaying the highest correlation coefficient with Composite Physiologic Index (CPI, r = 0.612). Multiple linear regression results indicated honeycombing was the best predictor for both the Dlco% and the CPI. Furthermore, multivariable logistic regression analysis identified honeycombing as an independent risk factor for GAP staging. Honeycombing differentiated between GAP I and GAP II + III with the highest AUC value of 0.729 (95% confidence interval: 0.634-0.811). With a threshold set at 8.0%, a sensitivity of 79.3% and a specificity of 57.4% were observed. Significant differences in honeycombing and ILD% were also noted among the disease groups (P < 0.05).

CONCLUSIONS: An AA% of 9.7% was the optimal threshold for differentiating CTD-ILD patients from HCs. Honeycombing can preliminarily predict lung function impairment and was an independent risk factor for GAP staging, offering significant clinical guidance for assessing the severity of the patient's disease.

PMID:39761506 | DOI:10.1097/RCT.0000000000001693

J Comput Assist Tomogr. 2024 Nov 13. doi: 10.1097/RCT.0000000000001680. Online ahead of print.

ABSTRACT

OBJECTIVE: This preliminary study aims to assess the image quality of enhanced-resolution deep learning reconstruction (ER-DLR) in magnetic resonance cholangiopancreatography (MRCP) and compare it with non-ER-DLR MRCP images.

METHODS: Our retrospective study incorporated 34 patients diagnosed with biliary and pancreatic disorders. We obtained MRCP images using a single breath-hold MRCP on a 3T MRI system. We reconstructed MRCP images with ER-DLR (matrix = 768 × 960) and without ER-DLR (matrix = 256 × 320). Quantitative evaluation involved measuring the signal-to-noise ratio (SNR), contrast, contrast-to-noise ratio (CNR) between the common bile duct and periductal tissues, and slope. Two radiologists independently scored image noise, contrast, artifacts, sharpness, and overall image quality for the 2 image types using a 4-point scale. Results are expressed as median and interquartile range (IQR), and we compared quantitative and qualitative scores employing the Wilcoxon test.

RESULTS: In quantitative analyses, ER-DLR significantly improved SNR (21.08 [IQR: 14.85, 31.5] vs 15.07 [IQR: 9.57, 25.23], P < 0.001), CNR (19.29 [IQR: 13.87, 24.98] vs 11.23 [IQR: 8.98, 15.74], P < 0.001), contrast (0.96 [IQR: 0.94, 0.97] vs 0.9 [IQR: 0.87, 0.92], P < 0.001), and slope of MRCP (0.62 [IQR: 0.56, 0.66] vs 0.49 [IQR: 0.45, 0.53], P < 0.001). The qualitative evaluation demonstrated significant improvements in the perceived noise (P < 0.001), contrast (P = 0.013), sharpness (P < 0.001), and overall image quality (P < 0.001).

CONCLUSIONS: ER-DLR markedly increased the resolution, SNR, and CNR of breath-hold-MRCP compared to cases without ER-DLR.

PMID:39761494 | DOI:10.1097/RCT.0000000000001680

PLoS Comput Biol. 2025 Jan 6;21(1):e1012744. doi: 10.1371/journal.pcbi.1012744. Online ahead of print.

ABSTRACT

Synonymous mutations, once considered neutral, are now understood to have significant implications for a variety of diseases, particularly cancer. It is indispensable to identify these driver synonymous mutations in human cancers, yet current methods are constrained by data limitations. In this study, we initially investigate the impact of sequence-based features, including DNA shape, physicochemical properties and one-hot encoding of nucleotides, and deep learning-derived features from pre-trained chemical molecule language models based on BERT. Subsequently, we propose EPEL, an effect predictor for synonymous mutations employing ensemble learning. EPEL combines five tree-based models and optimizes feature selection to enhance predictive accuracy. Notably, the incorporation of DNA shape features and deep learning-derived features from chemical molecule represents a pioneering effect in assessing the impact of synonymous mutations in cancer. Compared to existing state-of-the-art methods, EPEL demonstrates superior performance on independent test datasets. Furthermore, our analysis reveals a significant correlation between effect scores and patient outcomes across various cancer types. Interestingly, while deep learning methods have shown promise in other fields, their DNA sequence representations do not significantly enhance the identification of driver synonymous mutations in this study. Overall, we anticipate that EPEL will facilitate researchers to more precisely target driver synonymous mutations. EPEL is designed with flexibility, allowing users to retrain the prediction model and generate effect scores for synonymous mutations in human cancers. A user-friendly web server for EPEL is available at http://ahmu.EPEL.bio/.

PMID:39761306 | DOI:10.1371/journal.pcbi.1012744

PLoS One. 2025 Jan 6;20(1):e0313856. doi: 10.1371/journal.pone.0313856. eCollection 2025.

ABSTRACT

The consumption forecasting of oil and coal can help governments optimize and adjust energy strategies to ensure energy security in China. However, such forecasting is extremely challenging because it is influenced by many complex and uncertain factors. To fill this gap, we propose a hybrid deep learning approach for consumption forecasting of oil and coal in China. It consists of three parts, i.e., feature engineering, model building, and model integration. First, feature engineering is to distinguish the different correlations between targeted indicators and various features. Second, model building is to build five typical deep learning models with different characteristics to forecast targeted indicators. Third, model integration is to ensemble the built five models with a tailored, self-adaptive weighting strategy. As such, our approach enjoys all the merits of the five deep learning models (they have different learning structures and temporal constraints to diversify them for ensembling), making it able to comprehensively capture all the characteristics of different indicators to achieve accurate forecasting. To evaluate the proposed approach, we collected the real 880 pieces of data with 39 factors regarding the energy consumption of China ranging from 1999 to 2021. By conducting extensive experiments on the collected datasets, we have identified the optimal features for four targeted indicators (i.e., import of oil, production of oil, import of coal, and production of coal), respectively. Besides, we have demonstrated that our approach is significantly more accurate than the state-of-the-art forecasting competitors.

PMID:39761291 | DOI:10.1371/journal.pone.0313856

PLoS One. 2025 Jan 6;20(1):e0316642. doi: 10.1371/journal.pone.0316642. eCollection 2025.

ABSTRACT

OBJECTIVE: This study aimed to assess the feasibility of the deep learning in generating T2 weighted (T2W) images from diffusion-weighted imaging b0 images.

MATERIALS AND METHODS: This retrospective study included 53 patients who underwent head magnetic resonance imaging between September 1 and September 4, 2023. Each b0 image was matched with a corresponding T2-weighted image. A total of 954 pairs of images were divided into a training set with 763 pairs and a test set with 191 pairs. The Hybrid-Fusion Network (Hi-Net) and pix2pix algorithms were employed to synthesize T2W (sT2W) images from b0 images. The quality of the sT2W images was evaluated using three quantitative indicators: Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), and Normalized Mean Squared Error (NMSE). Subsequently, two radiologists were required to determine the authenticity of (s)T2W images and further scored the visual quality of sT2W images in the test set using a five-point Likert scale. The overall quality score, anatomical sharpness, tissue contrast and homogeneity were used to reflect the quality of the images at the level of overall and focal parts.

RESULTS: The indicators of pix2pix algorithm in test set were as follows: PSNR, 20.549±1.916; SSIM, 0.702±0.0864; NMSE, 0.239±0.150. The indicators of Hi-Net algorithm were as follows: PSNR, 20.646 ± 2.194; SSIM, 0.722 ± 0.0955; NMSE, 0.469 ± 0.124. Hi-Net performs better than pix2pix, so the sT2W images obtained by Hi-Net were used for radiologist assessment. The two readers accurately identified the nature of the images at rates of 69.90% and 71.20%, respectively. The synthetic images were falsely identified as real at rates of 57.6% and 57.1%, respectively. The overall quality score, sharpness, tissue contrast, and image homogeneity of the sT2Ws images ranged between 1.63 ± 0.79 and 4.45 ± 0.88. Specifically, the quality of the brain parenchyma, skull and scalp, and middle ear region was superior, while the quality of the orbit and paranasal sinus region was not good enough.

CONCLUSION: The Hi-Net is able to generate sT2WIs from low-resolution b0 images, with a better performance than pix2pix. It can therefore help identify incidental lesion through providing additional information, and demonstrates the potential to shorten the acquisition time of brain MRI during acute ischemic stroke imaging.

PMID:39761257 | DOI:10.1371/journal.pone.0316642

Radiol Phys Technol. 2025 Jan 6. doi: 10.1007/s12194-024-00874-y. Online ahead of print.

ABSTRACT

Self-supervised learning (SSL) has gained attention in the medical field as a deep learning approach utilizing unlabeled data. The Jigsaw puzzle task in SSL enables models to learn both features of images and the positional relationships within images. In breast cancer diagnosis, radiologists evaluate not only lesion-specific features but also the surrounding breast structures. However, deep learning models that adopt a diagnostic approach similar to human radiologists are still limited. This study aims to evaluate the effectiveness of the Jigsaw puzzle task in characterizing breast tissue structures for breast cancer classification on mammographic images. Using the Chinese Mammography Database (CMMD), we compared four pre-training pipelines: (1) IN-Jig, pre-trained with both the ImageNet classification task and the Jigsaw puzzle task, (2) Scratch-Jig, pre-trained only with the Jigsaw puzzle task, (3) IN, pre-trained only with the ImageNet classification task, and (4) Scratch, that is trained from random initialization without any pre-training tasks. All pipelines were fine-tuned using binary classification to distinguish between the presence or absence of breast cancer. Performance was evaluated based on the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Additionally, detailed analysis was conducted for performance across different radiological findings, breast density, and regions of interest were visualized using gradient-weighted class activation mapping (Grad-CAM). The AUC for the four models were 0.925, 0.921, 0.918, 0.909, respectively. Our results suggest the Jigsaw puzzle task is an effective pre-training method for breast cancer classification, with the potential to enhance diagnostic accuracy with limited data.

PMID:39760975 | DOI:10.1007/s12194-024-00874-y