T-stochastic Neighbor Embedding Research Articles

A study on classifying BSR noise in automotive door trim using a deep learning method

In this paper, we demonstrated a deep learning model that can automatically classifies BSR(Buzz, Squeak, Rattle) noise in automotive door trim using convolutional neural networks. In the preprocessing process, sound features of door trim noises are extracted by using STFT(Short-time Fourier Transform) algorithm to obtain spectrogram images for each noises we obtained from experiment. To classify various noises of door trim, those image data are labeled with its root cause, then it is trained with deep convolutional neural network. Furthermore, t-SNE (t-Stochastic Neighbor Embedding) algorithm was used to visualize the effect of deep convolutional neural network we demonstrated. The classifier of the demonstrated method achieved with the accuracy of 96%.

Comparison of the Effectiveness of Various Classifiers for Breast Cancer Detection Using Data Mining Methods

Countless women and men worldwide have lost their lives to breast cancer (BC). Although researchers from around the world have proposed various diagnostic methods for detecting this disease, there is still room for improvement in the accuracy and efficiency with which they can be used. A novel approach has been proposed for the early detection of BC by applying data mining techniques to the levels of prolactin (P), testosterone (T), cortisol (C), and human chorionic gonadotropin (HCG) in the blood and saliva of 20 women with histologically confirmed BC, 20 benign subjects, and 20 age-matched control women. In the proposed method, blood and saliva were used to categorize the severity of the BC into normal, benign, and malignant cases. Ten statistical features were collected to identify the severity of the BC using three different classification schemes—a decision tree (DT), a support vector machine (SVM), and k-nearest neighbors (KNN) were evaluated. Moreover, dimensionality reduction techniques using factor analysis (FA) and t-stochastic neighbor embedding (t-SNE) have been computed to obtain the best hyperparameters. The model has been validated using the k-fold cross-validation method in the proposed approach. Metrics for gauging a model’s effectiveness were applied. Dimensionality reduction approaches for salivary biomarkers enhanced the results, particularly with the DT, thereby increasing the classification accuracy from 66.67% to 93.3% and 90%, respectively, by utilizing t-SNE and FA. Furthermore, dimensionality reduction strategies for blood biomarkers enhanced the results, particularly with the DT, thereby increasing the classification accuracy from 60% to 80% and 93.3%, respectively, by utilizing FA and t-SNE. These findings point to t-SNE as a potentially useful feature selection for aiding in the identification of patients with BC, as it consistently improves the discrimination of benign, malignant, and control healthy subjects, thereby promising to aid in the improvement of breast tumour early detection.

A Novel Model Combining Circulating Absolute Monocyte Counts and a Four-Gene Monocyte Signature in Leukapheresis Identifies Lymphoma Patients at Very High Risk of Progression after Tisagenlecleucel or Axicabtagene Ciloleucel Therapy

Introduction CD19-directed chimeric antigen receptor (CAR) T can induce durable remissions in a fraction of relapsed/refractory large B-cell lymphomas (R/R LBCL), but 60% of patients (pts) still relapse. Biological mechanisms explaining lack of responses are emerging but they are largely unsuccessful in predicting disease response at the patient level. Aim To address the potential role of lymphocyte features before CAR T manufacturing, we performed transcriptomic and functional evaluations of leukapheresis products (LK) in 95 R/R LBCL pts enrolled in a prospective observational study, to identify correlates of response and survival to tisagenlecleucel (Tisa-cel) and axicabtagene ciloleucel (Axi-cel). Methods RNA of sorted CD3+ cells from LK was digitally quantified using the nCounter CAR T Characterization Panel (NanoString). Flow cytometry (FCM) was used for CAR T enumeration and monocyte analysis. Disease response at day 90 was assessed by computed tomography (CT) and positron emission tomography-CT (PET/CT) according to Lugano criteria. Statistical analyses were performed by GraphPad Prism v.9.00 and R v.4.1.2. Results We identified a 4-gene transcriptional signature (including MS4A4A, CD86, CD163 and SIGLEC5) able to divide LBCL pts into two groups, [namely Expressors (EXP) and poor-Expressors (poor-EXP)], characterized by significantly different survival probabilities. EXP have shorter PFS (P<0.0001: median PFS for EXP was 61 days and not reached for poor-EXP) (Fig. 1) and are less likely to respond (P<0.01). As the identified genes are highly expressed in monocytes but not in T cells, we sought to define whether monocytes contained in LK have a role in determining post CAR T outcome. In line with the fact that EXP had higher frequencies of monocytes in LK by FCM, we confirmed elevated absolute monocyte counts (AMC) both in LK (P<0.01) and in peripheral blood (PB) (P<0.05) of EXP. We then checked if monocyte levels could influence response and survival. When compared to responders, non-responders (NR) had higher LK AMC (P<0.01), higher percentages of monocytes (P<0.01) but also decreased percentages of lymphocytes (P<0.001), a parameter that was also associated with reduced PFS [HR: 0.97 (CI 0.96 - 0.99) P=0.01]. Consistent with the fact that circulating monocytes significantly correlated with those in LK (r= 0.6621, P<0.0001), NR also had significantly higher AMC (P<0.001) and AMC levels in PB negatively correlated with PFS [HR: 1.8 (CI 1.2 - 2.9) P=0.01]. Additionally, pts with serum C-reactive protein (CRP) and ferritin levels higher than the upper normal limit, displayed significantly higher AMC (P<0.001 for CRP; P<0.05 for ferritin). Accordingly, univariate analysis indicated that high ferritin, CRP and LDH levels at the time of LK were associated with the lack of response, but only high serum ferritin levels correlated with shorter PFS [HR: 2.86 (CI 1.25-6.55) P=0.013]. We next investigated if the combination of the monocyte signature and high ferritin levels and/or high circulating monocytes could better identify a group of pts at high risk of progression, compared to the expression of the 4 genes alone. Remarkably, albeit high ferritin levels and monocyte counts are overall associated with significantly reduced survival, the shortest PFS was achieved in EXP with monocyte counts above the median (P=0.0234: median PFS for EXP with high monocytes was 32.5 days and for EXP with low monocytes 209 days) (Fig. 2), thus supporting the hypothesis that CAR T treatment outcome is influenced by the presence of elevated levels of monocytes expressing the 4 genes. Concatenated monocyte FCM data from 30 pts analyzed with FlowSOM, followed by a consensus clustering algorithm combined with t-SNE (t-stochastic neighbor embedding), indicated that EXP had higher frequencies of CD14hi monocytes co-expressing MS4A4A, CD86, CD163 and SIGLEC5 at the protein level than poor-EXP (P<0.05), and accordingly, the presence of this population was also associated with lack of response (P<0.05). Conclusions Our data suggest that the poor outcome of some LBCL pts receiving CAR T is due to the presence of high monocytes levels expressing a four-gene signature in an inflammatory microenvironment. Additionally, the pre-manufacturing combined analysis of the gene signature and of PB AMC, could be used to plan trials evaluating CAR T cells versus other newly approved therapies such as bispecific antibodies.

Transferable wind power probabilistic forecasting based on multi-domain adversarial networks

Due to the limited availability of historical data, forecasting the wind power of newly-built wind farms poses a significant challenge. Transfer learning methods offer a potential solution by transferring knowledge from the source domain to the target domain, thereby reducing the data requirements for wind power forecasting. However, the difference in dataset distribution between the source and target domains causes domain shift. To address this issue, we propose a multi-domain adversarial network (MDAN). MDAN uses multi-domain datasets for adversarial learning, which maps numerical weather prediction (NWP) data to an adaptive latent space, thereby reducing domain shift. Additionally, we propose a data fusion-based incremental learning method to mitigate catastrophic forgetting. Through comprehensive case studies, MDAN provides accurate short-term wind power probabilistic forecasts in zero-shot learning. The incremental learning method enhances forecast accuracy in few-shot learning. Moreover, visualization analysis using t-stochastic neighbor embedding (t-SNE) shows that MDAN successfully reduces the domain shift between the source and target domains.

Analysis of heartbeat time series via machine learning for detection of illnesses

The heart, a component of the cardiovascular system, is responsible for pumping oxygenated and deoxygenated blood. It does not behave like a metronome and normally there is a variation in the duration of the intervals between each heartbeat, called Heart Rate Variability (HRV). In the presence of diseases or with the progression of aging, there is a reduction in HRV due to dysfunction of the autonomic nervous system. The objective of this work is to show, using machine learning techniques, that these techniques are able to relate directly the variability of the heart with the degree of the disease. Producing, as a practical result, the use of these techniques in the prediction of different types of diseases by only analyzing their time series. One of the first techniques used in our work is the unsupervised learning algorithm (t-Stochastic Neighbor Embedding). We show that this algorithm is able to differentiate the type and degree of the disease just by analyzing time series, we demonstrate that it is possible to design a neural network architecture capable of learning these characteristics, relating cardiac variability and the disease. In a complementary analysis, we check that cardiac variability can be directly related to permutation entropy, proving that the healthier an individual is, the more stochastic his cardiac time series is. We build a classification algorithm, using deep learning, from the confusion matrix and the ROC curve. This algorithm can be used as an entry point in diagnosing patients by measuring their HRV.

Multilingual Customized Keyword Spotting Using Similar-Pair Contrastive Learning

The goal of keyword spotting (KWS) is to find words and voice commands. KWS research has received a lot of attention recently. However, the majority of them emphasize predefined KWS. More often than not, customers wish to choose individualized unique keywords. In this work, we present a brand-new Similar-Pair Contrastive Learning (SPCL) training method called MCKWS-SPCL to address Multilingual Customized Keyword Spotting. The three key processes in our MCKWS-SPCL are acoustic feature extraction, embedding model, and similarity computation. By pushing similar-pair samples closer together in projection space, the SPCL training approach is specifically created to obtain similar features with noise existing. Additionally, multiple embedding neural network structures are carefully constructed to examine the trade-off between a limited number of parameters and excellent performance in order to reduce power consumption. For the first time, we complete an all-around multi-dimensional comparison to demonstrate the superior performance of our SPCL over conventional Contrastive Learning in KWS. Results from all trials, including metrics for average accuracy(96.84%), the detection error tradeoff curve, the false reject rate under 5% false alarm rate per hour(1.901%), and visualization using t-Stochastic Neighbor Embedding(same class is closer than CL), conclusively demonstrate our framework's effectiveness and superiority to Contrastive Learning-based approaches. Additionally, the analysis of four distinct language datasets and implementation on the NVIDIA JETSON TX2 hardware platform demonstrates the versatility, affordability, and hardware-friendliness of our new approach.

Parametric t-stochastic neighbor embedding with quantum neural network

t-stochastic neighbor embedding (t-SNE) is a nonparametric data visualization method in classical machine learning. It maps the data from the high-dimensional space into a low-dimensional space, especially a two-dimensional plane, while maintaining the relationship or similarities between the surrounding points. In t-SNE, the initial position of the low-dimensional data is randomly determined, and the visualization is achieved by moving the low-dimensional data to minimize a cost function. Its variant called parametric t-SNE uses neural networks for this mapping. In this paper, we propose to use quantum neural networks for parametric t-SNE to reflect the characteristics of high-dimensional quantum data on low-dimensional data. We use fidelity-based metrics instead of Euclidean distance in calculating high-dimensional data similarity. To verify our proposed method, we visualize both classical (Iris dataset) and quantum (time-depending Hamiltonian dynamics) data for classification tasks. Since this method allows us to represent a quantum dataset in a higher dimensional Hilbert space by a quantum dataset in a lower dimension while keeping their similarity, the proposed method can also be used to compress quantum data for further quantum machine learning.

Visualizing Structures in Financial Time-Series Datasets through Affinity-Based Diffusion Transition Embedding

In this work, the author proposes a modified version of PHATE, a diffusion map-based embedding algorithm that is tuned for working on financial time-series data primarily. The new algorithm, financial affinity-based diffusion transition embedding (FATE), takes in user-specified distance metrics that make sense for time-series data and uses symmetrized f-divergences applied to the diffusion probabilities as the final embedding distance before passing them into a metric multidimensional scaling step. The proposed visualization method reveals both local and global structures of the input time-series dataset. Performance of this visualization algorithm is first demonstrated through numerical experiments with Dow Jones 30 stock returns and S&P 100 stock returns. The author compares FATE visualization results using correlation-type distances with t-stochastic neighbor embedding and PHATE embeddings, among others, to demonstrate the advantages and new perspectives of FATE both qualitatively and quantitatively. On the other hand, experiments on synthetic ARMA time series with fine control of the structure of the underlying model parameters are provided. The results demonstrate the ability of transfer function information distance and time-lagged Hellinger distance to identify structures within the generating time-series models from their time-series realizations alone, which cannot be identified by correlation-type distances or Euclidean distances. The author concludes that the choice of distance metrics has an important role in the kind of structure one can uncover from time-series datasets.

Nivolumab-Based Salvage Therapy to Restore T Cell Fitness in Penta-Refractory Multiple Myeloma with Relapse to Anti-BCMA CAR T Cell Therapy

Background: Idecabtagene Vicleucel (ide-cel) is the first FDA-approved BCMA-directed CAR T cell therapy for the treatment of relapsed/refractory multiple myeloma (RRMM). Despite remarkable activity of 85% objective response rate and a progression-free survival (PFS) of 11.8 months reported in the CRB-401 trial (NCT02658929), most responses to ide-cel are not durable, and nearly all patients (pts) eventually relapse, even those achieving MRD-negative complete remission (CR). Factors associated with clinical relapse to ide-cel include BCMA loss, lack of CAR T persistence, T-cell exhaustion and PD-1/ PD-L1 impaired T-cell fitness. Here, we report clinical data and correlative findings for 4 pts who were salvaged with a nivolumab (anti-PD1)- based triplet therapy at the time of clinical relapse to first or second ide-cel infusion. Methods: Four pts with penta-refractory RRMM were treated as part of the CRB-401 phase 1 study as previously reported (Raje et al.N Engl J Med.2019;380:1726-37). All pts received nivolumab along with lenalidomide-dexamethasone (Nivo-Rd) or pomalidomide-dexamethasone (Nivo-Pd) at the time of relapse to ide-cel. CD3+ T cells (n=1,047) were isolated from the peripheral blood of 2 of these pts and full-length single-cell RNA sequencing (scRNA-seq) was conducted before and upon treatment with Nivo-Rd/-Pd. Response rates were assessed as per IMWG criteria. Results: The median age of pts was 48.5 years (range 37-56). Two out of 4 pts had cytogenetic high-risk MM and all pts were penta-refractory at the time of CRB-401 study inclusion. The median PFS and OS after first ide-cel infusion were 10.4 (range 3.0-13.8) months and 21.5 (range 16.0-34.6) months, respectively. Three pts received a second ide-cel infusion, whereof only 1 pt achieved a partial response (PR) at a PFS of 5.2 months. Two pts each were treated with Nivo-Rd and Nivo-Pd respectively at the time of progression to ide-cel. Responses to anti-PD1- based therapy varied markedly with 1 PR, 1 stable disease (SD) and 2 progressive disease (PD) as per IMWG criteria. The median PFS to nivolumab-based salvage therapy was 3.6 (range 0.4-8.1) months. Interestingly, pts MM1 and MM2 both had a CR to first ide-cel infusion (PFS 13.8 and 11.9 months, respectively), but while both pts did not respond to second CAR T cell infusion, the benefit from anti-PD1 treatment differed markedly in both pts. MM1 reached a PR and PFS of 105 days. In contrast, MM2 experienced immediate PD and succumbed to his disease 3.0 months later. Profiling of CD3+ T cells from MM1 and MM2 by scRNA-seq revealed rapid cellular state changes in response to PD1 inhibition. PAGODA2 clustering and t-Stochastic Neighbor Embedding (t-SNE) visualization revealed 3 distinct clusters, segregating to an individual pre-anti-PD1 cluster for MM1 and MM2 each and a shared post-anti-PD1 cluster comprising CD3+ T cells from both pts. The pre-anti-PD1 cluster for MM1 contained CD3+ T-cells with a memory T cell- like phenotype (IL7Rhigh, CD62Lhigh) and decreased expression of markers correlating with T-cell fitness (GNLY, PRF1, NKG7, GZMH,LY6C, KLRG1). This phenotype could be rescued, and T-cell fitness could be restored by Nivo-Pd treatment, potentially contributing to the clinical response observed in MM1. Opposingly, the pre-anti-PD1 cluster for MM2 showed intact T-cell fitness with low expression of T-cell exhaustion markers. No clinical benefit to Nivo-Rd was noted in this pt. Conclusions: Checkpoint inhibition and IMiD therapy can induce clinical responses at the time of relapse to first or second CAR T infusion. Efficacy however seems to depend on impaired T-cell fitness as premise for PD1 inhibition to re-induce clinical responses in selected pts.

Systematic investigation of keywords selection and processing strategy on search engine forecasting: a case of tourist volume in Beijing

The timeliness, precision, and low cost of search data have great potential for projecting tourist volume. Obtaining valuable information for decision-making, particularly for predicting, is hampered by the vast amount of search data. A systematic investigation of keyword selection and processing has been conducted. Using Beijing tourist volume as an example, 11 different feature extraction algorithms were selected and combined with long short-term memory (LSTM), random forest (RF) and fuzzy time series (FTS) for forecasting tourist volume. A total of 1612 keywords were retrieved from Baidu Index demand mapping using the direct word extraction method, range word extraction method and empirical selection method. The remaining 813 keywords were subjected to feature extraction. Based on the forecasting results of medium and short-term (1-day, 7-days and 10-days), the forecasting results of Kernel principal component analysis (KPCA) and locally linear embedding (LLE) are relatively stable when the dimensionality is reduced to 5 dimensions. The forecasting results of t-stochastic neighbor embedding (t-SNE), isometric mapping (IsoMap) and locally linear embedding (LLE), locality preserving projections (LPP), independent component correlation (ICA) are relatively stable when the dimensionality is reduced to 10 dimensions. Accurately forecasting many factors (transportation, attraction, food, lodging, travel, tips, tickets, and weather) provides a solid foundation for tourism demand optimization and scientific management and a resource for tourists' holistic vacation planning.

Fibronectin-1 is a dominant mechanism for rheumatoid arthritis via the mediation of synovial fibroblasts activity.

Rheumatoid arthritis (RA) has a high incidence and adverse effects on patients, thus posing a serious threat to people’s life and health. However, the underlying mechanisms regarding the development of RA are still elusive. Herein, we aimed to evaluate the RA-associated molecular mechanisms using the scRNA-seq technique. We used the GEO database to obtain scRNA-seq datasets for synovial fibroblasts (SFs) from RA cases, and the genes were then analyzed using principal component analysis (PCA) and T-Stochastic Neighbor Embedding (TSNE) analyses. Bioinformatics evaluations were carried out for asserting the highly enriched signaling pathways linked to the marker genes, and the key genes related to RA initiation were further identified. According to the obtained results, 3 cell types (0, 1, and 2) were identified by TSNE and some marker genes were statistically upregulated in cell type 1 than the other cell types. These marker genes predominantly contributed to extracellular matrix (ECM) architecture, collagen-harboring ECM, and ECM structural components, and identified as enriched with PI3K/AKT signaling cascade. Notably, fibronectin-1 (FN-1) has been identified as a critical gene that is strongly linked to the development of SFs and has enormous promise for regulating the onset of RA. Moreover, such an investigation offers novel perspectives within onset/progression of RA, suggesting that FN-1 may be a key therapeutic target for RA therapies.

Deep Learning Approach for Detecting Work-Related Stress Using Multimodal Signals

Work-related stress causes serious negative physiological and socioeconomic effects on employees. Detecting stress levels in a timely manner is important for appropriate stress management; therefore, this study proposes a deep learning (DL) approach that accurately detects work-related stress by using multimodal signals. We designed a protocol that simulates stressful situations and recruited 24 subjects for the experiments. Then, we collected electrocardiogram (ECG), respiration (RESP), and video data. The datasets were pre-processed and 10-s ECG and RESP signals and a sequence of facial features were fed into our deep neural network. Sixty-eight facial landmarks’ coordinates were extracted, and facial textures were extracted from a pre-trained network based on facial expression recognition. Each signal was processed by each of its network branch, and data were fused at two different levels: 1) feature-level and 2) decision-level. The feature-level fusion that used RESP and facial landmarks’ coordinates showed average accuracy of 73.3%, AUC of 0.822, and F1 score of 0.700 in two-level stress classification, and the feature-level fusion that used ECG, RESP, and the coordinates showed average accuracy of 54.4%, AUC of 0.727, and F1 score of 0.508 in three-level stress classification. When analyzing the weights in the decision-level fusion, we found that the importance of each information item varied according to the stress classification problem. When comparing t-stochastic neighbor embedding results, we observed that overlapped samples of different classes caused performance degradation in both classifications. Our findings suggest that the proposed DL approach fusing multimodal and heterogeneous signals can enhance stress detection.

An unsupervised machine learning approach to evaluate sports facilities condition in primary school.

Sports facilities have been acknowledged as one of the crucial environmental factors for children’s physical education, physical fitness, and participation in physical activity. Finding a solution for the effective and objective evaluation of the condition of sports facilities in schools (SSFs) with the responding quantitative magnitude is an uncertain task. This paper describes the utilization of an unsupervised machine learning method to objectively evaluate the condition of sports facilities in primary school (PSSFC). The statistical data of 845 samples with nine PSSFC indicators (indoor and outdoor included) were collected from the Sixth National Sports Facility Census in mainland China (NSFC), an official nationwide quinquennial census. The Fuzzy C-means (FCM) algorithm was applied to cluster the samples in accordance with the similarity of PSSFC. The clustered data were visualized by using t-stochastic neighbor embedding (t-SNE). The statistics results showed that the application of t-SNE and FCM led to the acceptable performance of clustering SSFs data into three types with differences in PSSFC. The effects of school category, location factors, and the interaction on PSSFC were analyzed by two-way analysis of covariance, which indicated that regional PSSFC has geographical and typological characteristics: schools in the suburbs are superior to those in the inner city, schools with more grades of students are configured with better variety and larger size of sports facilities. In conclusion, we have developed a combinatorial machine learning clustering approach that is suitable for objective evaluation on PSSFC and indicates its characteristics.

Industrial process data visualization based on a deep enhanced t-distributed stochastic neighbor embedding neural network

PurposeThe purpose of this paper is to propose a approach for data visualization and industrial process monitoring.Design/methodology/approachA deep enhanced t-distributed stochastic neighbor embedding (DESNE) neural network is proposed for data visualization and process monitoring. The DESNE is composed of two deep neural networks: stacked variant auto-encoder (SVAE) and a deep label-guided t-stochastic neighbor embedding (DLSNE) neural network. In the DESNE network, SVAE extracts informative features of the raw data set, and then DLSNE projects the extracted features to a two dimensional graph.FindingsThe proposed DESNE is verified on the Tennessee Eastman process and a real data set of blade icing of wind turbines. The results indicate that DESNE outperforms some visualization methods in process monitoring.Originality/valueThis paper has significant originality. A stacked variant auto-encoder is proposed for feature extraction. The stacked variant auto-encoder can improve the separation among classes. A deep label-guided t-SNE is proposed for visualization. A novel visualization-based process monitoring method is proposed.

Non-Parametric Semi-Supervised Learning in Many-Body Hilbert Space with Rescaled Logarithmic Fidelity

In quantum and quantum-inspired machine learning, a key step is to embed the data in the quantum space known as Hilbert space. Studying quantum kernel function, which defines the distances among the samples in the Hilbert space, belongs to the fundamental topics in this direction. In this work, we propose a tunable quantum-inspired kernel function (QIKF) named rescaled logarithmic fidelity (RLF) and a non-parametric algorithm for the semi-supervised learning in the quantum space. The rescaling takes advantage of the non-linearity of the kernel to tune the mutual distances of samples in the Hilbert space, and meanwhile avoids the exponentially-small fidelities between quantum many-qubit states. Being non-parametric excludes the possible effects from the variational parameters, and evidently demonstrates the properties of the kernel itself. Our results on the hand-written digits (MNIST dataset) and movie reviews (IMDb dataset) support the validity of our method, by comparing with the standard fidelity as the QIKF as well as several well-known non-parametric algorithms (naive Bayes classifiers, k-nearest neighbors, and spectral clustering). High accuracy is demonstrated, particularly for the unsupervised case with no labeled samples and the few-shot cases with small numbers of labeled samples. With the visualizations by t-stochastic neighbor embedding, our results imply that the machine learning in the Hilbert space complies with the principles of maximal coding rate reduction, where the low-dimensional data exhibit within-class compressibility, between-class discrimination, and overall diversity. The proposed QIKF and semi-supervised algorithm can be further combined with the parametric models such as tensor networks, quantum circuits, and quantum neural networks.

Accuracy of one-step automated orthodontic diagnosis model using a convolutional neural network and lateral cephalogram images with different qualities obtained from nationwide multi-hospitals.

ObjectiveThe purpose of this study was to investigate the accuracy of one-step automated orthodontic diagnosis of skeletodental discrepancies using a convolutional neural network (CNN) and lateral cephalogram images with different qualities from nationwide multi-hospitals.MethodsAmong 2,174 lateral cephalograms, 1,993 cephalograms from two hospitals were used for training and internal test sets and 181 cephalograms from eight other hospitals were used for an external test set. They were divided into three classification groups according to anteroposterior skeletal discrepancies (Class I, II, and III), vertical skeletal discrepancies (normodivergent, hypodivergent, and hyperdivergent patterns), and vertical dental discrepancies (normal overbite, deep bite, and open bite) as a gold standard. Pre-trained DenseNet-169 was used as a CNN classifier model. Diagnostic performance was evaluated by receiver operating characteristic (ROC) analysis, t-stochastic neighbor embedding (t-SNE), and gradient-weighted class activation mapping (Grad-CAM).ResultsIn the ROC analysis, the mean area under the curve and the mean accuracy of all classifications were high with both internal and external test sets (all, > 0.89 and > 0.80). In the t-SNE analysis, our model succeeded in creating good separation between three classification groups. Grad-CAM figures showed differences in the location and size of the focus areas between three classification groups in each diagnosis.ConclusionsSince the accuracy of our model was validated with both internal and external test sets, it shows the possible usefulness of a one-step automated orthodontic diagnosis tool using a CNN model. However, it still needs technical improvement in terms of classifying vertical dental discrepancies.

A Novel Computational Framework for Precision Diagnosis and Subtype Discovery of Plant With Lesion.

Plants are often attacked by various pathogens during their growth, which may cause environmental pollution, food shortages, or economic losses in a certain area. Integration of high throughput phenomics data and computer vision (CV) provides a great opportunity to realize plant disease diagnosis in the early stage and uncover the subtype or stage patterns in the disease progression. In this study, we proposed a novel computational framework for plant disease identification and subtype discovery through a deep-embedding image-clustering strategy, Weighted Distance Metric and the t-stochastic neighbor embedding algorithm (WDM-tSNE). To verify the effectiveness, we applied our method on four public datasets of images. The results demonstrated that the newly developed tool is capable of identifying the plant disease and further uncover the underlying subtypes associated with pathogenic resistance. In summary, the current framework provides great clustering performance for the root or leave images of diseased plants with pronounced disease spots or symptoms.

Joint t-SNE for Comparable Projections of Multiple High-Dimensional Datasets.

We present Joint t-Stochastic Neighbor Embedding (Joint t-SNE), a technique to generate comparable projections of multiple high-dimensional datasets. Although t-SNE has been widely employed to visualize high-dimensional datasets from various domains, it is limited to projecting a single dataset. When a series of high-dimensional datasets, such as datasets changing over time, is projected independently using t-SNE, misaligned layouts are obtained. Even items with identical features across datasets are projected to different locations, making the technique unsuitable for comparison tasks. To tackle this problem, we introduce edge similarity, which captures the similarities between two adjacent time frames based on the Graphlet Frequency Distribution (GFD). We then integrate a novel loss term into the t-SNE loss function, which we call vector constraints, to preserve the vectors between projected points across the projections, allowing these points to serve as visual landmarks for direct comparisons between projections. Using synthetic datasets whose ground-truth structures are known, we show that Joint t-SNE outperforms existing techniques, including Dynamic t-SNE, in terms of local coherence error, Kullback-Leibler divergence, and neighborhood preservation. We also showcase a real-world use case to visualize and compare the activation of different layers of a neural network.

Insight into the potential pathogenesis of human osteoarthritis via single-cell RNA sequencing data on osteoblasts.

Osteoarthritis (OA) is the most common degenerative joint disease caused by osteoblastic lineage cells. However, a comprehensive molecular program for osteoblasts in human OA remains underdeveloped. The single-cell gene expression of osteoblasts and microRNA array data were from human. After processing the single-cell RNA sequencing (scRNA-seq) data, it was subjected to principal component analysis (PCA) and T-Stochastic neighbor embedding analysis (TSNE). Differential expression analysis was aimed to find marker genes. Gene-ontology (GO) enrichment, Kyoto encyclopedia of genes and genomes (KEGG) pathway enrichment analysis and Gene set enrichment analysis (GSEA) were applied to characterize the molecular function of osteoblasts with marker genes. Protein-protein interaction (PPI) networks and core module were established for marker genes by using the STRING database and Cytoscape software. All nodes in the core module were considered to be hub genes. Subsequently, we predicted the potential miRNA of hub genes through the miRWalk, miRDB and TargetScan database and experimentally verified the miRNA by GSE105027. Finally, miRNA-mRNA regulatory network was constructed using the Cytoscape software. We characterized the single-cell expression profiling of 4387 osteoblasts from normal and OA sample. The proportion of osteoblasts subpopulations changed dramatically in the OA, with 70.42% of the pre-osteoblasts. 117 marker genes were included and the results of GO analysis show that up-regulated marker genes enriched in collagen-containing extracellular matrix were highly expressed in the pre-osteoblasts cluster. Both KEGG and GSEA analyses results indicated that IL-17 and NOD-like receptor signaling pathways were enriched in down-regulated marker genes. We visualize the weight of marker genes and constructed the core module in PPI network. In potential mRNA-miRNA regulatory network, hsa-miR-449a and hsa-miR-218-5p may be involved in the development of OA. Our study found that alterations in osteoblasts state and cellular molecular function in the subchondral bone region may be involved in the pathogenesis of osteoarthritis.

Depth Prototype Clustering Method Based on Unsupervised Field Alignment for Bearing Fault Identification of Mechanical Equipment

In the process of mechanical equipment fault diagnosis, it is difficult to obtain enough labeled samples due to the changeable operating conditions, complex working environment, and the limitation of measuring equipment. Therefore, a prototype clustering method based on unsupervised domain alignment is proposed for fault detection and diagnosis of mechanical equipment under unmarked samples. First, the relative maximum mean discrepancy index is constructed to map the data collected from different devices to the same high-dimensional space for feature alignment. Then, cluster the output and calculate the prototype for each category. Compared with the classification directly using the softmax function, the prototype clustering method can correct the local possible misjudged points through the overall average distribution. By minimizing the global distance error, prototypes obtained by clustering in the target domain are matched with prototypes of tagged samples in the source domain and, finally, realize unsupervised state recognition. Six experiments are designed to verify the effectiveness of the proposed method. The accuracy of the proposed method is 1%–2% higher even when compared to one of the most advanced fault diagnosis methods available. Also, it can be found from the results of t-stochastic neighbor embedding (SNE) visualization that the proposed method has higher knowledge transfer capability.