Network. 2024 Dec 29:1-30. doi: 10.1080/0954898X.2024.2447281. Online ahead of print.

ABSTRACT

To improve the calculation accuracy of the Monte Carlo (MC) method and reduce the calculation time. Firstly, CNN and LSTM deep learning networks are introduced for designing nonlinear dynamic systems simulating dam stress. Then, spatial feature mining and sequence information extraction of nonlinear data of dam stress are carried out respectively, and a combined prediction model of dam stress depth (DS-FEM-CNN-LSTM) is proposed. Secondly, to solve the problem of a long time and heavy workload for the MC method to calculate a single sample point, the DOE test method is used to design the sample points. The weight factor and the distance to the failure surface are used as screening criteria. The reliability calculation method of the gravity dam (DS-FEM-CNN-LSTM-MC) is established. Finally, numerical results show that the proposed DS-FEM-CNN-LSTM-MC method performs better than the existing methods in terms of computational time consumption and accuracy.

PMID:39733444 | DOI:10.1080/0954898X.2024.2447281

Asian Pac J Cancer Prev. 2024 Dec 1;25(12):4201-4210. doi: 10.31557/APJCP.2024.25.12.4201.

ABSTRACT

BACKGROUND: Glioblastoma (GBM) is the most aggressive cancer in the central nervous system in glial cells. Finding novel biomarkers in GBM offers numerous advantages that can contribute to early detection, personalized treatment, improved patient outcomes, and advancements in cancer research and drug development. Integrating machine learning with RNAseq data in medicine holds significant potential for identifying novel biomarkers in various diseases, including cancer.

METHODS: Gene expression raw data was used to detect differentially expressed genes (DEGs) within a cohort of 532 GBM patients. The molecular pathway analysis, disease ontology, and protein-protein interactions of DEGs were assessed. Machine learning methods were performed to identify candidate genes. Survival curves were estimated using the Kaplan-Meier method and Cox proportional hazard to find prognostic biomarkers.

RESULTS: The molecular pathway analysis revealed that key dysregulated genes are in GPCRs, class A rhodopsin-like, MAPK signaling pathway, and calcium regulation in cardiac cells. Additionally, survival analysis showed that ten downregulated genes, including CPLX3, GPR162, LCNL1, SLC5A5, GPR61, GPR68, IL1RL2, HCRTR1, AIPL1, and SYTL1, and also ten upregulated genes, including C1orf92, CATSPER1, CCDC19, EPS8L1, FAIM3, FAM70B, FCN3, GPR157, IGFBP1, and MYBPH decreased the overall survival in GBM patients. Furthermore, the machine learning detected twenty genes, among which LRRTM2 and OPRL1 were candidates with high correlation coefficients.

CONCLUSION: Our data suggest that genes belonging to G Protein-Coupled Receptors play a critical role in various aspects of glioblastoma progression and pathogenesis. Four members of GPCRs, including GPR162, GPR61, GPR68, and GPR157, can be considered prognostic biomarkers. Additionally, the combination of A2BP1 and GPR157 was reported as a diagnostic marker.

PMID:39733410 | DOI:10.31557/APJCP.2024.25.12.4201

Sci Rep. 2024 Dec 28;14(1):31426. doi: 10.1038/s41598-024-83088-x.

ABSTRACT

The scattering of tiny particles in the atmosphere causes a haze effect on remote sensing images captured by satellites and similar devices, significantly disrupting subsequent image recognition and classification. A generative adversarial network named TRPC-GAN with texture recovery and physical constraints is proposed to mitigate this impact. This network not only effectively removes haze but also better preserves the texture information of the original remote sensing image, thereby enhancing the visual quality of the dehazed image. A multi-scale module is proposed to extract feature information of remote sensing images, allowing it to capture image features from different receptive fields. Simultaneously, an attention module is designed further to guide the network's focus towards important feature information. In addition, a multi-scale adversarial network is proposed to better restore both global and local information about the original image. Introducing a physical constraint loss function to improve the loss function of the original generative adversarial network allows for better preservation of the physical characteristics of remote sensing images. Simulation experiments on synthetic and natural hazy remote sensing image datasets are conducted. The results demonstrate that the dehazing performance of the TRPC-GAN method surpasses the other four methods.

PMID:39733123 | DOI:10.1038/s41598-024-83088-x

Sci Rep. 2024 Dec 28;14(1):31479. doi: 10.1038/s41598-024-83347-x.

ABSTRACT

This study aimed to explore a deep learning radiomics (DLR) model based on grayscale ultrasound images to assist radiologists in distinguishing between benign breast lesions (BBL) and malignant breast lesions (MBL). A total of 382 patients with breast lesions were included, comprising 183 benign lesions and 199 malignant lesions that were collected and confirmed through clinical pathology or biopsy. The enrolled patients were randomly allocated into two groups: a training cohort and an independent test cohort, maintaining a ratio of 7:3.We created a model called CLDLR that utilizes clinical parameters and DLR to diagnose both BBL and MBL through grayscale ultrasound images. In order to assess the practicality of the CLDLR model, two rounds of evaluations were conducted by radiologists. The CLDLR model demonstrates the highest diagnostic performance in predicting benign and malignant BI-RADS 4 lesions, with areas under the receiver operating characteristic curve (AUC) of 0.988 (95% confidence interval : 0.949, 0.985) in the training cohort and 0.888 (95% confidence interval : 0.829, 0.947) in the testing cohort.The CLDLR model outperformed the diagnoses made by the three radiologists in the initial assessment of the testing cohorts. By utilizing AI scores from the CLDLR model and heatmaps from the DLR model, the diagnostic performance of all radiologists was further enhanced in the testing cohorts. Our study presents a noninvasive imaging biomarker for the prediction of benign and malignant BI-RADS 4 lesions. By comparing the results from two rounds of assessment, our AI-assisted diagnostic tool demonstrates practical value for radiologists with varying levels of experience.

PMID:39733121 | DOI:10.1038/s41598-024-83347-x

Sci Rep. 2024 Dec 28;14(1):31392. doi: 10.1038/s41598-024-82931-5.

ABSTRACT

Failure to predict stroke promptly may lead to delayed treatment, causing severe consequences like permanent neurological damage or death. Early detection using deep learning (DL) and machine learning (ML) models can enhance patient outcomes and mitigate the long-term effects of strokes. The aim of this study is to compare these models, exploring their efficacy in predicting stroke. This study analyzed a dataset comprising 663 records from patients hospitalized at Hazrat Rasool Akram Hospital in Tehran, Iran, including 401 healthy individuals and 262 stroke patients. A total of eight established ML (SVM, XGB, KNN, RF) and DL (DNN, FNN, LSTM, CNN) models were utilized to predict stroke. Techniques such as 10-fold cross-validation and hyperparameter tuning were implemented to prevent overfitting. The study also focused on interpretability through Shapley Additive Explanations (SHAP). The evaluation of model's performance was based on accuracy, specificity, sensitivity, F1-score, and ROC curve metrics. Among DL models, LSTM showed superior sensitivity at 96.15%, while FNN exhibited better specificity (96.0%), accuracy (96.0%), F1-score (95.0%), and ROC (98.0%) among DL models. For ML models, RF displayed higher sensitivity (99.9%), accuracy (99.0%), specificity (100%), F1-score (99.0%), and ROC (99.9%). Overall, RF outperformed all models, while DL models surpassed ML models in most metrics except for RF. DL models (CNN, LSTM, DNN, FNN) achieved sensitivities from 93.0 to 96.15%, specificities from 80.0 to 96.0%, accuracies from 92.0 to 96.0%, F1-scores from 87.34 to 95.0%, and ROC scores from 95.0 to 98.0%. In contrast, ML models (KNN, XGB, SVM) showed sensitivities between 29.0% and 94.0%, specificities between 89.47% and 96.0%, accuracies between 71.0% and 95.0%, F1-scores between 44.0% and 95.0%, and ROC scores between 64.0% and 95.0%. This study demonstrates the efficacy of DL and ML models in predicting stroke, with the RF models outperforming all others in key metrics. While DL models generally surpassed ML models, RF's exceptional performance highlights the potential of combining these technologies for early stroke detection, significantly improving patient outcomes by preventing severe consequences like permanent neurological damage or death.

PMID:39733046 | DOI:10.1038/s41598-024-82931-5

Sci Rep. 2024 Dec 28;14(1):31503. doi: 10.1038/s41598-024-82981-9.

ABSTRACT

The study aims to address the critical issue of toxic side effects resulting from drug combinations, which can significantly increase health risks, clinical complications, and lead to drug being withdrawn from the market. A model named TSEDDI (toxic side effects of drug-drug interaction) has been developed to improve the identification of drug pairs that may induce toxicity or adverse reactions. By utilizing drug chemical structures and diverse proteins, we employ a convolutional neural network (CNN) to extract features from molecular images, enzyme proteins, transporter proteins, and target proteins. Furthermore, we introduce a weighted binary cross entropy loss function to tackle class imbalance and integrate the multi-head attention mechanism with residual connections to enhance model performance. Our model outperformed advanced baseline models in predicting drug-drug interaction (DDI) side effects, achieving an accuracy of 0.9059 (± 0.0010) and consistently excelling across various evaluation metrics. The case study confirms the potential mechanisms by which four pairs of drugs cause side effects, thus demonstrating the effectiveness of our model in predicting DDI side effects. The TSEDDI model combines multiple attention mechanisms and residual connections, enhancing its ability to detect toxic and adverse effects related to DDIs. As a result, it becomes a valuable resource for promptly identifying adverse reactions in clinical trials. Future research could investigate drug substructures prone to toxic side effects.

PMID:39733005 | DOI:10.1038/s41598-024-82981-9

Sci Rep. 2024 Dec 28;14(1):31252. doi: 10.1038/s41598-024-82607-0.

ABSTRACT

Lithofacies classification and identification are of great significance in the exploration and evaluation of tight sandstone reservoirs. Existing methods of lithofacies identification in tight sandstone reservoirs face issues such as lengthy manual classification, strong subjectivity of identification, and insufficient sample datasets, which make it challenging to analyze the lithofacies characteristics of these reservoirs during oil and gas exploration. In this paper, the Fuyu oil formation in the Songliao Basin is selected as the target area, and an intelligent method for recognizing the lithophysics reservoirs in tight sandstone based on hybrid multilayer perceptron (MLP) and multivariate time series (MTS-Mixers) is proposed. Firstly, appropriate logging curve parameters are selected based on the contribution rate as the basis of the lithofacies intelligent discrimination. Second, preprocessing operations are performed on the logging dataset to ensure the quality of the experimental data. Finally, the MLP-MTS hybrid intelligent model is constructed by combining the powerful information extraction and classification recognition capabilities of the MLP and MTS models to complete the intelligent recognition of the petrography of tight sandstone reservoirs. The experimental results demonstrate that the recognition efficiency of MLP-MTS model for all kinds of lithofacies phases is more than 90%, which verifies the good applicability of deep learning model in solving the process of lithofacies phase recognition in reservoirs.

PMID:39732990 | DOI:10.1038/s41598-024-82607-0

Sci Rep. 2024 Dec 28;14(1):31269. doi: 10.1038/s41598-024-82681-4.

ABSTRACT

The expressway green channel is an essential transportation policy for moving fresh agricultural products in China. In order to extract knowledge from various records, this study presents a cutting-edge approach to extract information from textual records of failure cases in the vertical field of expressway green channel. We proposed a hybrid approach based on BIO labeling, pre-trained model, deep learning and CRF to build a named entity recognition (NER) model with the optimal prediction performance. Eight entities are designed and proposed in the NER processing for the expressway green channel. three typical pre-trained natural language processing models are utilized and compared to recognize entities and obtain feature vectors, including bidirectional encoder representations from transformer (BERT), ALBERT, and RoBERTa. An ablation experiment is performed to analyze the influence of each factor on the proposed models. Used the survey data from the expressway green channel management system in Shaanxi Province of China, the experimental results show that the precision, recall, and F1-score of the RoBERTa-BiGRU-CRF model are 93.04%, 92.99%, and 92.99%, respectively. As the results, it is discovered that the text features extracted from pre-training substantially enhance the prediction accuracy of deep learning algorithms. Surprisingly, the RoBERTa model is highly effective in the task for the expressway green channel NER. This study provides a timely and necessary knowledge extraction on the Expressway Green Channel in terms of textual data, offering a systematical explanation of failure cases and valuable insights for future research.

PMID:39732976 | DOI:10.1038/s41598-024-82681-4

Sci Rep. 2024 Dec 28;14(1):31409. doi: 10.1038/s41598-024-83028-9.

ABSTRACT

To improve students' understanding of physical education teaching concepts and help teachers analyze students' cognitive patterns, the study proposes an association learning-based method for understanding physical education teaching concepts using deep learning algorithms, which extracts image features related to teaching concepts using convolutional neural networks. Moreover, a neurocognitive diagnostic model based on hypergraph convolution is constructed to mine the data of students' long-term learning sequences and identify students' cognitive outcomes. The findings revealed that the highest accuracy of the association graph convolutional neural network was 0.84 when the number of training samples was 90,000. In each of the three datasets, the cognitive diagnostic model's accuracy was 0.76, 0.77, and 0.75, respectively. The use of the association graph convolutional neural network model resulted in an increase of 29% in the mastery of students in the concepts and knowledge of sports. The predictive accuracy of the cognitive schema diagnostic model ranged from 0.6 to 1.0 with a mean value of 0.81. The study reveals that the model proposed in the study has high accuracy and stability in predicting cognitive patterns, which can better identify students' cognitive states and provide strong support for instructional guidance and personalized learning.

PMID:39732971 | DOI:10.1038/s41598-024-83028-9

Sci Rep. 2024 Dec 28;14(1):31180. doi: 10.1038/s41598-024-82583-5.

ABSTRACT

Classification and regression problems can be challenging when the relevant input features are diluted in noisy datasets, in particular when the sample size is limited. Traditional Feature Selection (FS) methods address this issue by relying on some assumptions such as the linear or additive relationship between features. Recently, a proliferation of Deep Learning (DL) models has emerged to tackle both FS and prediction at the same time, allowing non-linear modeling of the selected features. In this study, we systematically assess the performance of DL-based feature selection methods on synthetic datasets of varying complexity, and benchmark their efficacy in uncovering non-linear relationships between features. We also use the same settings to benchmark the reliability of gradient-based feature attribution techniques for Neural Networks (NNs), such as Saliency Maps (SM). A quantitative evaluation of the reliability of these approaches is currently missing. Our analysis indicates that even simple synthetic datasets can significantly challenge most of the DL-based FS and SM methods, while Random Forests, TreeShap, mRMR and LassoNet are the best performing FS methods. Our conclusion is that when quantifying the relevance of a few non linearly-entangled predictive features diluted in a large number of irrelevant noisy variables, DL-based FS and SM interpretation methods are still far from being reliable.

PMID:39732866 | DOI:10.1038/s41598-024-82583-5

Sci Rep. 2024 Dec 28;14(1):31277. doi: 10.1038/s41598-024-82612-3.

ABSTRACT

Traditional hydraulic structures rely on manual visual inspection for apparent integrity, which is not only time-consuming and labour-intensive but also inefficient. The efficacy of deep learning models is frequently constrained by the size of available data, resulting in limited scalability and flexibility. Furthermore, the paucity of data diversity leads to a singular function of the model that cannot provide comprehensive decision support for improving maintenance measures. This paper proposes an efficacious methodology for identifying diverse apparent damages in hydraulic structures to address the limitations of existing technologies. The advanced features of apparent damage in hydraulic structures were elucidated by fine-tuning the top-level parameters of the lightweight pre-trained model, thereby mitigating the data dependency issue inherent in the model. Ensemble learning algorithms are employed to classify high-dimensional samples to enhance the accuracy and stability of the classification. However, ensemble learning algorithms are subject to time consuming issues when applied to high-dimensional datasets. To this end, we propose a robust discriminative feature selection model to identify the most salient features, thereby enhancing the performance of apparent damage recognition in hydraulic structures while concurrently reducing the inference time. The results demonstrated that the accuracies of this method in identifying crack, fracture, hole and normal structures were 87.65%, 87.82%, 96.99%, and 95.25%, respectively. This methodology exhibits significant applicability and practical value for the intelligent inspection of hydraulic structures.

PMID:39732863 | DOI:10.1038/s41598-024-82612-3

Sci Rep. 2024 Dec 28;14(1):31257. doi: 10.1038/s41598-024-82676-1.

ABSTRACT

Surface electromyography (sEMG) data has been extensively utilized in deep learning algorithms for hand movement classification. This paper aims to introduce a novel method for hand gesture classification using sEMG data, addressing accuracy challenges seen in previous studies. We propose a U-Net architecture incorporating a MobileNetV2 encoder, enhanced by a novel Bidirectional Long Short-Term Memory (BiLSTM) and metaheuristic optimization for spatial feature extraction in hand gesture and motion recognition. Bayesian optimization is employed as the metaheuristic approach to optimize the BiLSTM model's architecture. To address the non-stationarity of sEMG signals, we employ a windowing strategy for signal augmentation within deep learning architectures. The MobileNetV2 encoder and U-Net architecture extract relevant features from sEMG spectrogram images. Edge computing integration is leveraged to further enhance innovation by enabling real-time processing and decision-making closer to the data source. Six standard databases were utilized, achieving an average accuracy of 90.23% with our proposed model, showcasing a 3-4% average accuracy improvement and a 10% variance reduction. Notably, Mendeley Data, BioPatRec DB3, and BioPatRec DB1 surpassed advanced models in their respective domains with classification accuracies of 88.71%, 90.2%, and 88.6%, respectively. Experimental results underscore the significant enhancement in generalizability and gesture recognition robustness. This approach offers a fresh perspective on prosthetic management and human-machine interaction, emphasizing its efficacy in improving accuracy and reducing variance for enhanced prosthetic control and interaction with machines through edge computing integration.

PMID:39732856 | DOI:10.1038/s41598-024-82676-1

BMC Res Notes. 2024 Dec 28;17(1):393. doi: 10.1186/s13104-024-07062-6.

ABSTRACT

This dataset contains demographic, morphological and pathological data, endoscopic images and videos of 191 patients with colorectal polyps. Morphological data is included based on the latest international gastroenterology classification references such as Paris, Pit and JNET classification. Pathological data includes the diagnosis of the polyps including Tubular, Villous, Tubulovillous, Hyperplastic, Serrated, Inflammatory and Adenocarcinoma with Dysplasia Grade & Differentiation.Objectives: Today the most important challenge of developing accurate algorithms for medical prediction, detection, diagnosis, treatment and prognosis is data. ERCPMP is an Endoscopic Image and Video Dataset for Recognition of Colorectal Polyps Morphology and Pathology. This dataset can be used for developing deep learning algorithms for polyps detection, classification, and segmentation.Data description: Images were captured with Olympus colonoscope and are presented in RGB format, JPG type with the resolution of 368 * 256 pixels and 96 dpi. The name of each file (image or video) includes pathological diagnosis, grade and JNet classification of the related polyp.

PMID:39732672 | DOI:10.1186/s13104-024-07062-6

BMC Genomics. 2024 Dec 28;25(1):1253. doi: 10.1186/s12864-024-11168-3.

ABSTRACT

Long non-coding RNAs (lncRNAs) play crucial roles in numerous biological processes and are involved in complex human diseases through interactions with proteins. Accurate identification of lncRNA-protein interactions (LPI) can help elucidate the functional mechanisms of lncRNAs and provide scientific insights into the molecular mechanisms underlying related diseases. While many sequence-based methods have been developed to predict LPIs, efficiently extracting and effectively integrating potential feature information that reflects functional attributes from lncRNA and protein sequences remains a significant challenge. This paper proposes a Dinucleotide-Codon Fusion Feature encoding (DNCFF) and constructs an LPI prediction model based on deep learning, termed LPI-DNCFF. The Dual Nucleotide Visual Fusion Feature encoding (DNVFF) incorporates positional information of single nucleotides with subsequent nucleotide connections, while Codon Fusion Feature encoding (CFF) considers the specificity, molecular weight, and physicochemical properties of each amino acid. These encoding methods encapsulate rich and intuitive sequence information in limited encoding dimensions. The model comprehensively predicts LPIs by integrating global, local, and structural features, and inputs them into BiLSTM and attention layers to form a hybrid deep learning model. Experimental results demonstrate that LPI-DNCFF effectively predicts LPIs. The BiLSTM layer and attention mechanism can learn long-term dependencies and identify weighted key features, enhancing model performance. Compared to one-hot encoding, DNCFF more efficiently and thoroughly extracts potential sequence features. Compared to other existing methods, LPI-DNCFF achieved the best performance on the RPI1847 and ATH948 datasets, with MCC values of approximately 97.84% and 84.58%, respectively, outperforming the state-of-the-art method by about 1.44% and 3.48%.

PMID:39732642 | DOI:10.1186/s12864-024-11168-3

Acad Radiol. 2024 Dec 27:S1076-6332(24)00894-8. doi: 10.1016/j.acra.2024.11.049. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: To evaluate and compare image quality of different energy levels of virtual monochromatic images (VMIs) using standard versus strong deep learning spectral reconstruction (DLSR) on dual-energy CT pulmonary angiogram (DECT-PA).

MATERIALS AND METHODS: A retrospective study was performed on 70 patients who underwent DECT-PA (15 PE present; 55 PE absent) scans. VMIs were reconstructed at different energy levels ranging from 35 to 200 keV using standard and strong levels with deep learning spectral reconstruction. Quantitative assessment was performed using region of interest (ROI) analysis of eleven different anatomical areas, measuring absolute attenuation, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). In addition, CNR of clot compared to normally opacified lumen was calculated in cases that were positive for PE. For qualitative analysis, four different keV levels (40-60-80-100) were evaluated.

RESULTS: The image noise was significantly lower, and the cardiovascular SNR (24.9 ± 5.85 vs. 21.98 ± 5.49) and CNR (23.72 ± 8.00 vs. 20.31 ± 6.44) were significantly higher, on strong Deep Learning Spectral reconstruction (DLSR) than standard DLSR (p < 0.0001). PE-specific CNR (8.58 ± 4.47 vs. 6.25 ± 3.19) was significantly higher on strong DLSR than standard (p < 0.0001). The subjective image quality scores were diagnostically acceptable at four different keV levels (40-60-80-100 keV) evaluated using both standard and strong DLSR, with no qualitative differences observed at those energies.

CONCLUSION: Strong DLSR improves image quality with an increase of the SNR and CNR in DECT-PA compared to standard DLSR.

PMID:39732618 | DOI:10.1016/j.acra.2024.11.049

Am J Pathol. 2024 Dec 26:S0002-9440(24)00469-3. doi: 10.1016/j.ajpath.2024.10.023. Online ahead of print.

ABSTRACT

Understanding the tumor hypoxic microenvironment is crucial for grasping tumor biology, clinical progression, and treatment responses. This study presents a novel application of AI in computational histopathology to evaluate hypoxia in breast cancer. Weakly Supervised Deep Learning (WSDL) models can accurately detect morphological changes associated with hypoxia in routine Hematoxylin and Eosin (H&E) whole slide images (WSI). Our model, HypOxNet, was trained on H&E WSI from breast cancer primary sites (n=1016) at 40x magnification using data from The Cancer Genome Atlas (TCGA). We utilized the Hypoxia Buffa signature to measure hypoxia scores, which ranged from -43 to +47, and stratified the samples into hypoxic and normoxic based on these scores. This stratification represented the weak labels associated with each WSI. HypOxNet achieved an average Area Under the Curve (AUC) of 0.82 on test sets, identifying significant differences in cell morphology between hypoxic and normoxic tissue regions. Importantly, once trained, the HypOxNet model requires only the readily available H&E slides, making it especially valuable in low-resource settings where additional gene expression assays are not available. These AI-based hypoxia detection models can potentially be extended to other tumor types and seamlessly integrated into pathology workflows, offering a fast, cost-effective alternative to molecular testing.

PMID:39732389 | DOI:10.1016/j.ajpath.2024.10.023

Neurobiol Dis. 2024 Dec 26:106783. doi: 10.1016/j.nbd.2024.106783. Online ahead of print.

ABSTRACT

DYT-THAP1 dystonia is a monogenetic form of dystonia, a movement disorder characterized by the involuntary co-contraction of agonistic and antagonistic muscles. The disease is caused by mutations in the THAP1 gene, although the precise mechanisms by which these mutations contribute to the pathophysiology of dystonia remain unclear. The incomplete penetrance of DYT-THAP1 dystonia, estimated at 40 to 60 %, suggests that an environmental trigger may be required for the manifestation of the disease in genetically predisposed individuals. To investigate the gene-environment interaction in the development of dystonic features, we performed a sciatic nerve crush injury in a genetically predisposed DYT-THAP1 heterozygous knockout mouse model (Thap1+/-). We employed a multi-omic assessment to study the pathophysiological pathways underlying the disease. Phenotypic analysis using an unbiased deep learning algorithm revealed that nerve-injured Thap1+/- mice exhibited significantly more dystonia like movements (DLM) over the course of the 12-week experiment compared to naive Thap1+/- mice. In contrast, nerve-injured wildtype (wt) mice only showed a significant increase in DLM compared to their naive counterpart during the first weeks after injury. Furthermore, at week 11 after nerve crush, nerve-injured Thap1+/- mice displayed significantly more DLM than nerve-injured wt counterparts. Multi-omic analysis of the cerebellum, striatum and cortex in nerve-injured Thap1+/- mice revealed differences that are indicative of an altered energy metabolism compared to naive Thap1+/- and nerve-injured wt animals. These findings suggest that aberrant energy metabolism in brain regions relevant to dystonia may underlie the dystonic phenotype observed in nerve injured Thap1+/- mice.

PMID:39732371 | DOI:10.1016/j.nbd.2024.106783

Lab Invest. 2024 Dec 26:102221. doi: 10.1016/j.labinv.2024.102221. Online ahead of print.

ABSTRACT

Tumor cell nuclear size (NS) indicates malignant potential in breast cancer; however, its clinical significance in esophageal squamous cell carcinoma (ESCC) is unknown. Artificial intelligence (AI) can quantitatively evaluate histopathological findings. The aim was to measure NS in ESCC using AI and elucidate its clinical significance. We investigated the relationship between NS assessed by AI and prognosis in 138 patients with ESCC who underwent curative esophagectomy. Hematoxylin and eosin-stained slides from the deepest tumor sections were digitized. Using HALO-AI DenseNet v2, we created a deep learning classifier that identified tumor cells with an NS area >20 μm2. Median NS was 40.14 μm2, which was used to divide patients into NS-high and NS-low groups (n = 69 per group). Five-year overall survival (OS) and relapse-free survival rates were significantly lower in the NS-high group (43.2% and 39.6%) than in the NS-low group (67.7% and 49.6%). Multivariate analysis showed that greater tumor depth and NS-high status (hazard ratio [HR]: 1.79; p = 0.032) were independent risk factors for OS. In 77 cases with neoadjuvant chemotherapy, increased tumor depth and NS-high status (HR: 1.99; p = 0.048) were independent prognostic factors for unfavorable OS. Compared to the NS-low group, the NS-high group had significantly higher anisokaryosis, higher Ki-67 expression as calculated by AI analysis of immunostaining, and higher NS heterogeneity as examined by equidividing the tumors into square tiles. In conclusion, NS assessed by AI is a simple and useful prognostic factor for ESCC.

PMID:39732367 | DOI:10.1016/j.labinv.2024.102221

Neural Netw. 2024 Dec 20;184:107053. doi: 10.1016/j.neunet.2024.107053. Online ahead of print.

ABSTRACT

Deep learning systems are prone to catastrophic forgetting when learning from a sequence of tasks, as old data from previous tasks is unavailable when learning a new task. To address this, some methods propose replaying data from previous tasks during new task learning, typically using extra memory to store replay data. However, it is not expected in practice due to memory constraints and data privacy issues. Instead, data-free replay methods invert samples from the classification model. While effective, these methods face inconsistencies between inverted and real training data, overlooked in recent works. To that effect, we propose to measure the data consistency quantitatively by some simplification and assumptions. Using this measurement, we gain insight to develop a novel loss function that reduces inconsistency. Specifically, the loss minimizes the KL divergence between distributions of inverted and real data under a tied multivariate Gaussian assumption, which is simple to implement in continual learning. Additionally, we observe that old class weight norms decrease continually as learning progresses. We analyze the reasons and propose a regularization term to balance class weights, making old class samples more distinguishable. To conclude, we introduce Consistency-enhanced data replay with a Debiased classifier for class incremental learning (CwD). Extensive experiments on CIFAR-100, Tiny-ImageNet, and ImageNet100 show consistently improved performance of CwD compared to previous approaches.

PMID:39732067 | DOI:10.1016/j.neunet.2024.107053

Comput Biol Med. 2024 Dec 27;186:109629. doi: 10.1016/j.compbiomed.2024.109629. Online ahead of print.

ABSTRACT

BACKGROUND: Studying influential nodes (I-nodes) in brain networks is of great significance in the field of brain imaging. Most existing studies consider brain connectivity hubs as I-nodes such as the regions of high centrality or rich-club organization. However, this approach relies heavily on prior knowledge from graph theory, which may overlook the intrinsic characteristics of the brain network, especially when its architecture is not fully understood. In contrast, self-supervised deep learning dispenses with manual features, allowing it to learn meaningful representations directly from the data. This approach enables the exploration of I-nodes for brain networks, which is also lacking in current studies.

METHOD: This paper proposes a Self-Supervised Graph Reconstruction framework based on Graph-Transformer (SSGR-GT) to identify I-nodes, which has three main characteristics. First, as a self-supervised model, SSGR-GT extracts the importance of brain nodes to the reconstruction. Second, SSGR-GT uses Graph-Transformer, which is well-suited for extracting features from brain graphs, combining both local and global characteristics. Third, multimodal analysis of I-nodes uses graph-based fusion technology, combining functional and structural brain information.

RESULTS: The I-nodes we obtained are distributed in critical areas such as the superior frontal lobe, lateral parietal lobe, and lateral occipital lobe, with a total of 56 identified across different experiments. These I-nodes are involved in more brain networks than other regions, have longer fiber connections, and occupy more central positions in structural connectivity. They also exhibit strong connectivity and high node efficiency in both functional and structural networks. Furthermore, there is a significant overlap between the I-nodes and both the structural and functional rich-club.

CONCLUSIONS: Experimental results verify the effectiveness of the proposed method, and I-nodes are obtained and discussed. These findings enhance our understanding of the I-nodes within the brain network, and provide new insights for future research in further understanding the brain working mechanisms.

PMID:39731922 | DOI:10.1016/j.compbiomed.2024.109629