Deep Least Squares Fisher Discriminant Analysis

Development and Validation of Chinese Version of Dental Pain Screening Questionnaire

INTRODUCTION

Dental pain is one of the most prevalent oral complaints. This study aimed to establish a Chinese version of Dental Pain Screening Questionnaire (DePaQ) to help classify patients into three groups of dental pain diseases (group 1, reversible pulpitis and dentine hypersensitivity; group 2, acute periapical periodontitis and irreversible pulpitis; and group 3, pericoronitis).

METHODS

The DePaQ was translated from English to Chinese. The clinical subjects (CS, n = 290) and nonclinical subjects (NS, n = 100) with dental pain were collected. The CS were randomly divided into two subsamples: CS1 (n = 203) and CS2 (n = 87). The Fisher discriminant functions of the final 13-item Chinese version of the DePaQ were obtained from the CS1 group, and discriminant validity was further verified in the CS2 and NS groups.

RESULTS

The discriminant functions of the final 13-item DePaQ obtained from the CS1 group were capable of correctly distinguishing 93.1% and 89.0% cases of the CS2 and NS groups, respectively. In the CS2 group, the sensitivity for groups 1, 2, and 3 was 88.0%, 80.0%, and 83.0%, respectively, and the specificity was 95.0%, 95.0%, and 86.0%, respectively. In the NS group, the sensitivity for groups 1, 2, and 3 was 82.0%, 80.0%, and 86.0%, respectively, and the specificity was 91.0%, 97.0%, and 90.0%, respectively.

CONCLUSIONS

The Chinese version of DePaQ could help classify the three groups of dental pain diseases and guide medical treatment.

Immune subtyping of melanoma whole slide images using multiple instance learning

Determining early-stage prognostic markers and stratifying patients for effective treatment are two key challenges for improving outcomes for melanoma patients. Previous studies have used tumour transcriptome data to stratify patients into immune subgroups, which were associated with differential melanoma specific survival and potential predictive biomarkers. However, acquiring transcriptome data is a time-consuming and costly process. Moreover, it is not routinely used in the current clinical workflow. Here, we attempt to overcome this by developing deep learning models to classify gigapixel haematoxylin and eosin (H&E) stained pathology slides, which are well established in clinical workflows, into these immune subgroups. We systematically assess six different multiple instance learning (MIL) frameworks, using five different image resolutions and three different feature extraction methods. We show that pathology-specific self-supervised models using 10x resolution patches generate superior representations for the classification of immune subtypes. In addition, in a primary melanoma dataset, we achieve a mean area under the receiver operating characteristic curve (AUC) of 0.80 for classifying histopathology images into 'high' or 'low immune' subgroups and a mean AUC of 0.82 in an independent TCGA melanoma dataset. Furthermore, we show that these models are able to stratify patients into 'high' and 'low immune' subgroups with significantly different melanoma specific survival outcomes (log rank test, P< 0.005). We anticipate that MIL methods will allow us to find new biomarkers of high importance, act as a tool for clinicians to infer the immune landscape of tumours and stratify patients, without needing to carry out additional expensive genetic tests.

DisP+V: A Unified Framework for Disentangling Prototype and Variation From Single Sample per Person

Single sample per person face recognition (SSPP FR) is one of the most challenging problems in FR due to the extreme lack of enrolment data. To date, the most popular SSPP FR methods are the generic learning methods, which recognize query face images based on the so-called prototype plus variation (i.e., P+V) model. However, the classic P+V model suffers from two major limitations: 1) it linearly combines the prototype and variation images in the observational pixel-spatial space and cannot generalize to multiple nonlinear variations, e.g., poses, which are common in face images and 2) it would be severely impaired once the enrolment face images are contaminated by nuisance variations. To address the two limitations, it is desirable to disentangle the prototype and variation in a latent feature space and to manipulate the images in a semantic manner. To this end, we propose a novel disentangled prototype plus variation model, dubbed DisP+V, which consists of an encoder-decoder generator and two discriminators. The generator and discriminators play two adversarial games such that the generator nonlinearly encodes the images into a latent semantic space, where the more discriminative prototype feature and the less discriminative variation feature are disentangled. Meanwhile, the prototype and variation features can guide the generator to generate an identity-preserved prototype and the corresponding variation, respectively. Experiments on various real-world face datasets demonstrate the superiority of our DisP+V model over the classic P+V model for SSPP FR. Furthermore, DisP+V demonstrates its unique characteristics in both prototype recovery and face editing/interpolation.

Discriminant Feature Extraction by Generalized Difference Subspace

In this paper, we reveal the discriminant capacity of orthogonal data projection onto the generalized difference subspace (GDS), both theoretically and experimentally. In our previous work, we demonstrated that the GDS projection works as a quasi-orthogonalization of class subspaces, which is an effective feature extraction for subspace based classifiers. Here, we further show that GDS projection also works as a discriminant feature extraction through a similar mechanism to the Fisher discriminant analysis (FDA). A direct proof of the connection between GDS projection and FDA is difficult due to the significant difference in their formulations. To circumvent the complication, we first introduce geometrical Fisher discriminant analysis (gFDA) based on a simplified Fisher criterion. It is derived from a heuristic yet practically plausible assumption: the direction of the sample mean vector of a class is largely aligned to the first principal component vector of the class, given that the principal component analysis (PCA) is applied without data centering. gFDA works stably even under few samples, bypassing the small sample size (SSS) problem of FDA. We then prove that gFDA is equivalent to GDS projection with a small correction term. This equivalence ensures GDS projection to inherit the discriminant ability from FDA via gFDA. Furthermore, we discuss two useful extensions of these methods, 1) a nonlinear extension by kernel trick, 2) a combination with CNN features. The equivalence and the effectiveness of the extensions have been verified through extensive experiments on the extended Yale B+, CMU face database, ALOI, ETH80, MNIST, and CIFAR10, mainly focusing on image recognition under small samples.

Novel variation mode decomposition integrated adaptive sparse principal component analysis and it application in fault diagnosis

The sparse principal component analysis (SPCA) is widely used in the fault detection for nonlinear complex chemical processes in recent years. However, insufficient data processing, fixed models and fault type single classification cannot be used in the time-varying process. Therefore, a novel adaptive sparse principal component analysis (ASPCA) algorithm fused with improved variation mode decomposition (IVMD) (ASPCA-IVMD) is proposed for fault detection in chemical processes. The bat algorithm is innovatively integrated to optimize the parameters of the variable modulus decomposition. Then the optimized parameters are used for data preprocessing to suppress noise. In addition, based on the traditional SPCA, the threshold calculation is fused to realize the adaptive selection of principal components. After the principal components are determined, T² and Q statistics are used for fault detection. Finally, the proposed method is verified by the Tennessee Eastman process case. The results demonstrate that the proposed method can select the principal components adaptively according to the data for having the real-time property of chemical process. Meanwhile, compared with traditional methods (principal component analysis, sparse principal component analysis, deep belief network integrating dropout, adaptive unscented Kalman filter integrating radial basis function and sparse deep belief network), the detection rate of the ASPCA-IVMD method is more than 99%, which shows superiority.

Prediction of carotid plaque by blood biochemical indices and related factors based on Fisher discriminant analysis

OBJECTIVE

This study aims to establish the predictive model of carotid plaque formation and carotid plaque location by retrospectively analyzing the clinical data of subjects with carotid plaque formation and normal people, and to provide technical support for screening patients with carotid plaque.

METHODS

There were 4300 subjects in the ultrasound department of Maanshan People's Hospital collected from December 2013 to December 2018. We used demographic and biochemical data from 3700 subjects to establish predictive models for carotid plaque and its location. The leave-one-out cross-validated classification, 600 external data validation, and area under the receiver operating characteristic curve (AUC) were used to verify the accuracy, sensitivity, specificity, and application value of the model.

RESULTS

There were significant difference of age (F = - 34.049, p < 0.01), hypertension (χ² = 191.067, p < 0.01), smoking (χ² = 4.762, p < 0.05) and alcohol (χ² = 8.306, p < 0.01), Body mass index (F = 15.322, p < 0.01), High-density lipoprotein (HDL) (F = 13.840, p < 0.01), Lipoprotein a (Lp a) (F = 52.074, p < 0.01), Blood Urea Nitrogen (F = 2.679, p < 0.01) among five groups. Prediction models were built: carotid plaque prediction model (Model CP); Prediction model of left carotid plaque only (Model CP Left); Prediction model of right carotid plaque only (Model CP Right). Prediction model of bilateral carotid plaque (Model CP Both). Model CP (Wilks' lambda = 0.597, p < 0.001, accuracy = 78.50%, sensitivity = 78.07%, specificity = 79.07%, AUC = 0.917). Model CP Left (Wilks' lambda = 0.605, p < 0.001, accuracy = 79.00%, sensitivity = 86.17%, specificity = 72.70%, AUC = 0.880). Model CP Right (Wilks' lambda = 0.555, p < 0.001, accuracy = 83.00%, sensitivity = 81.82%, specificity = 84.44%, AUC = 0.880). Model CP Both (Wilks' lambda = 0.651, p < 0.001, accuracy = 82.30%, sensitivity = 89.50%, specificity = 72.70%, AUC = 0.880).

CONCLUSION

Demographic characteristics and blood biochemical indexes were used to establish the carotid plaque and its location discriminant models based on Fisher discriminant analysis (FDA), which has high application value in community screening.

Epoch-Evolving Gaussian Process Guided Learning for Classification

The conventional mini-batch gradient descent algorithms are usually trapped in the local batch-level distribution information, resulting in the ``zig-zag'' effect in the learning process. To characterize the correlation information between the batch-level distribution and the global data distribution, we propose a novel learning scheme called epoch-evolving Gaussian process guided learning (GPGL) to encode the global data distribution information in a non-parametric way. Upon a set of class-aware anchor samples, our GP model is built to estimate the class distribution for each sample in mini-batch through label propagation from the anchor samples to the batch samples. The class distribution, also named the context label, is provided as a complement for the ground-truth one-hot label. Such a class distribution structure has a smooth property and usually carries a rich body of contextual information that is capable of speeding up the convergence process. With the guidance of the context label and ground-truth label, the GPGL scheme provides a more efficient optimization through updating the model parameters with a triangle consistency loss. Furthermore, our GPGL scheme can be generalized and naturally applied to the current deep models, outperforming the state-of-the-art optimization methods on six benchmark datasets.

Discovering Panel of Autoantibodies for Early Detection of Lung Cancer Based on Focused Protein Array

Substantial studies indicate that autoantibodies to tumor-associated antigens (TAAbs) arise in early stage of lung cancer (LC). However, since single TAAbs as non-invasive biomarkers reveal low diagnostic performances, a panel approach is needed to provide more clues for early detection of LC. In the present research, potential TAAbs were screened in 150 serum samples by focused protein array based on 154 proteins encoded by cancer driver genes. Indirect enzyme-linked immunosorbent assay (ELISA) was used to verify and validate TAAbs in two independent datasets with 1,054 participants (310 in verification cohort, 744 in validation cohort). In both verification and validation cohorts, eight TAAbs were higher in serum of LC patients compared with normal controls. Moreover, diagnostic models were built and evaluated in the training set and the test set of validation cohort by six data mining methods. In contrast to the other five models, the decision tree (DT) model containing seven TAAbs (TP53, NPM1, FGFR2, PIK3CA, GNA11, HIST1H3B, and TSC1), built in the training set, yielded the highest diagnostic value with the area under the receiver operating characteristic curve (AUC) of 0.897, the sensitivity of 94.4% and the specificity of 84.9%. The model was further assessed in the test set and exhibited an AUC of 0.838 with the sensitivity of 89.4% and the specificity of 78.2%. Interestingly, the accuracies of this model in both early and advanced stage were close to 90%, much more effective than that of single TAAbs. Protein array based on cancer driver genes is effective in screening and discovering potential TAAbs of LC. The TAAbs panel with TP53, NPM1, FGFR2, PIK3CA, GNA11, HIST1H3B, and TSC1 is excellent in early detection of LC, and they might be new target in LC immunotherapy.