Improve Classification Performance Research Articles

Efficient deep learning-based automated diagnosis from echocardiography with contrastive self-supervised learning

BackgroundAdvances in self-supervised learning (SSL) have enabled state-of-the-art automated medical image diagnosis from small, labeled datasets. This label efficiency is often desirable, given the difficulty of obtaining expert labels for medical image recognition tasks. However, most efforts toward SSL in medical imaging are not adapted to video-based modalities, such as echocardiography.MethodsWe developed a self-supervised contrastive learning approach, EchoCLR, for echocardiogram videos with the goal of learning strong representations for efficient fine-tuning on downstream cardiac disease diagnosis. EchoCLR pretraining involves (i) contrastive learning, where the model is trained to identify distinct videos of the same patient, and (ii) frame reordering, where the model is trained to predict the correct of video frames after being randomly shuffled.ResultsWhen fine-tuned on small portions of labeled data, EchoCLR pretraining significantly improves classification performance for left ventricular hypertrophy (LVH) and aortic stenosis (AS) over other transfer learning and SSL approaches across internal and external test sets. When fine-tuning on 10% of available training data (519 studies), an EchoCLR-pretrained model achieves 0.72 AUROC (95% CI: [0.69, 0.75]) on LVH classification, compared to 0.61 AUROC (95% CI: [0.57, 0.64]) with a standard transfer learning approach. Similarly, using 1% of available training data (53 studies), EchoCLR pretraining achieves 0.82 AUROC (95% CI: [0.79, 0.84]) on severe AS classification, compared to 0.61 AUROC (95% CI: [0.58, 0.65]) with transfer learning.ConclusionsEchoCLR is unique in its ability to learn representations of echocardiogram videos and demonstrates that SSL can enable label-efficient disease classification from small amounts of labeled data.

Distinguishing between Wheat Grains Infested by Four Fusarium Species by Measuring with a Low-Cost Electronic Nose

An electronic device based on the detection of volatile substances was developed in response to the need to distinguish between fungal infestations in food and was applied to wheat grains. The most common pathogens belong to the fungi of the genus Fusarium: F. avenaceum, F. langsethiae, F. poae, and F. sporotrichioides. The electronic nose prototype is a low-cost device based on commercially available TGS series sensors from Figaro Corp. Two types of gas sensors that respond to the perturbation are used to collect signals useful for discriminating between the samples under study. First, an electronic nose detects the transient response of the sensors to a change in operating conditions from clean air to the presence of the gas being measured. A simple gas chamber was used to create a sudden change in gas composition near the sensors. An inexpensive pneumatic system consisting of a pump and a carbon filter was used to supply the system with clean air. It was also used to clean the sensors between measurement cycles. The second function of the electronic nose is to detect the response of the sensor to temperature disturbances of the sensor heater in the presence of the gas to be measured. It has been shown that features extracted from the transient response of the sensor to perturbations by modulating the temperature of the sensor heater resulted in better classification performance than when the machine learning model was built from features extracted from the response of the sensor in the gas adsorption phase. By combining features from both phases of the sensor response, a further improvement in classification performance was achieved. The E-nose enabled the differentiation of F. poae from the other fungal species tested with excellent performance. The overall classification rate using the Support Vector Machine model reached 70 per cent between the four fungal categories tested.

HClass: Fast hybrid network traffic classification with bit and keyword level signatures

Deep Packet Inspection (DPI) methods are extensively used in traffic classification. These methods extract unique application content either at byte or bit level granularity and represent them as signatures. DPI involves string or regular expression matching, which is computationally expensive, and evaluating signatures at bit-level granularity makes it even more inefficient. With the ever-increasing bandwidth and the high-speed internet traffic, the software implementations of DPI have become a performance bottleneck. In this paper, we propose HClass, a DPI-based network traffic classifier completely implemented in software to speed up signature matching. Our contributions with HClass are three-fold. First, we propose a hybrid signature matching technique with a combination of bit and byte-level signatures. Second, we propose methods to perform bit-level signature matching with byte/word level operations to cope with software implementations and be compatible with general-purpose CPU operations. Third, it uses a two-phase signature matching where first-phase signatures are short and quickly identify the potential application(s), and the second-phase signatures verify the potential application(s) to reduce false positives. We perform experiments with HClass on three datasets and report classification performance and execution time improvement of HClass with our implementations in C language.

Explainable feature selection and ensemble classification via feature polarity

Feature selection aims to choose the most relevant features from the dataset that can enhance the performance and efficiency of machine learning models. Although feature selection has been studied for many years, most existing methods focus on accuracy and efficiency while neglecting the interpretability of selected features. Therefore, inspired by the “Yin-Yang” philosophy, we introduce the concept of feature polarity for the first time and divide the features into positive and negative features. For example, by analyzing a patient's symptoms (features), we can obtain two sets of features to explain whether the patient has the flu. Positive features help us determine if the patient has the flu, while negative features can help us rule out the possibility of the flu. We introduce the PN (Positive and Negative) coefficient to measure the polarity of candidate features and develop a novel and explainable feature selection method based on feature polarity. Furthermore, we propose an ensemble classification framework that leverages both positive and negative features for each class to improve classification performance. Extensive experiments demonstrate the effectiveness of the PN coefficient compared to other information measurements. Moreover, our proposed classification framework performs excellently compared to some state-of-the-art feature selection methods.

Feature selection based on chi-square and ant colony optimization for multi-label classification

Text classification is widely used in organizations with large databases and digital documents. In text classification, there are many features, most of which are redundant. High-dimensional features impact multi-label classification performance. Feature selection is a data processing technique that can overcome this problem. Feature selection techniques have two major approaches: filter and wrapper. This paper proposes a hybrid filter-wrapper technique combining two algorithms: Chi-square (CS) and ant colony optimization (ACO). In the first stage, CS is used to reduce the number of irrelevant features. The ACO method is in the second stage. The ACO is applied to select the efficient features and improve classifier performance. The experiment results show that CS-ACO, CS-grey wolf optimizer (GWO), CS, and without feature selection (FS) have a micro F1-score based multinomial naïve Bayes classifier including 80%, 79.75%, 79.64% and 77.78%. The result indicates that the CS-ACO algorithm is suitable for solving multi-label classification problems.

Face mask classification using convolutional neural networks with facial image regions and super resolution

<p>Face mask classification is relevant to public health and safety, so an approach for face mask classification using Multi-Task Cascaded Convolutional Networks (MTCNN) for face detection on image data, ResNet152 architecture for feature extraction, and super-resolution method, BSRGAN, for enhanced image quality was proposed. The classification model was trained by a fully connected layer of neural networks. The goal is to classify each facial image into three classes: the image with a mask, without a mask, or with an incorrectly worn mask. The performance of each classification model on two real-world datasets was evaluated by Accuracy, Precision, Recall, and F1 score for different sets of input patterns which were features extracted from the facial image regions including their combinations. Using multiple image regions, i.e. face, nose, and mouth, as resources for preparing input features showed the improved classification performance compared to using single image regions. In addition, the super-resolution technique applied to medium or large-sized images can improve the performance of the face mask classification model. Our findings may further guide the development for greater effective models and techniques on face mask classification contributing to practical scenarios.</p>

Enhancing liver fibrosis diagnosis and treatment assessment: a novel biomechanical markers-based machine learning approach

Accurate diagnosis and treatment assessment of liver fibrosis face significant challenges, including inherent limitations in current techniques like sampling errors and inter-observer variability. Addressing this, our study introduces a novel machine learning (ML) framework, which integrates light gradient boosting machine and multivariate imputation by chained equations to enhance liver status assessment using biomechanical markers. Building upon our previously established multiscale mechanical characteristics in fibrotic and treated livers, this framework employs Gaussian Bayesian optimization for post-imputation, significantly improving classification performance. Our findings indicate a marked increase in the precision of liver fibrosis diagnosis and provide a novel, quantitative approach for assessing fibrosis treatment. This innovative combination of multiscale biomechanical markers with advanced ML algorithms represents a transformative step in liver disease diagnostics and treatment evaluation, with potential implications for other areas in medical diagnostics.

Classification of mango disease using ensemble convolutional neural network

Mango is a highly significant fruit crop that thrives in a variety of agro-ecologies around the world. Mangoes are rich in vitamins and minerals. However, its yield is currently severely constrained due to disease and pest infestations. Thus, in order to improve mango fruit quality and productivity, illnesses and insect pests must be detected early on. In this study, we conceived and constructed a mango leaf disease detection mechanism utilizing an ensemble convolutional neural network approach. Healthy and diseased mango leaf images were manually obtained from main producing locations in Amhara Region for Merawi fruit and vegetable research identification. To improve the datasets, several pre-processing procedures (such as image resizing, noise reduction, and image augmentation) were used. To improve classification performance and meet the study's purpose, various segmentation approaches such as k means and Mask R-CNN were applied. Furthermore, following pre-processing and segmentation, features of mango leaf images were retrieved using CNN to obtain important features. The classification model was then constructed using fully-connected layer classifiers on the retrieved features of mango leaf images. The ensemble proposed GoogLeNet and VGG16 based CNN model in the study encompasses various operations, including dataset collection, image preprocessing, noise removal, segmentation, data augmentation, feature extraction, and classification. Upon testing, the model demonstrated impressive performance with 99.87 % training classification accuracy, 99.72 % validation accuracy, and 99.21 % testing accuracy. This indicates the effectiveness of the ensemble approach in achieving high accuracy in image classification tasks.

DLA-Net: dual lesion attention network for classification of pneumoconiosis using chest X-ray images

Accurate and early detection of pneumoconiosis using chest X-rays (CXR) is important for preventing the progression of this incurable disease. It is also a challenging task due to large variations in appearance, size and location of lesions in the lung regions as well as inter-class similarity and intra-class variance. Compared to traditional methods, Convolutional Neural Networks-based methods have shown improved results; however, these methods are still not applicable in clinical practice due to limited performance. In some cases, limited computing resources make it impractical to develop a model using whole CXR images. To address this problem, the lung fields are divided into six zones, each zone is classified separately and the zone classification results are then aggregated into an image classification score, based on state-of-the-art. In this study, we propose a dual lesion attention network (DLA-Net) for the classification of pneumoconiosis that can extract features from affected regions in a lung. This network consists of two main components: feature extraction and feature refinement. Feature extraction uses the pre-trained Xception model as the backbone to extract semantic information. To emphasise the lesion regions and improve the feature representation capability, the feature refinement component uses a DLA module that consists of two sub modules: channel attention (CA) and spatial attention (SA). The CA module focuses on the most important channels in the feature maps extracted by the backbone model, and the SA module highlights the spatial details of the affected regions. Thus, both attention modules combine to extract discriminative and rich contextual features to improve classification performance on pneumoconiosis. Experimental results show that the proposed DLA-Net outperforms state-of-the-art methods for pneumoconiosis classification.

Nonparametric mean and variance adaptive classification rule for high‐dimensional data with heteroscedastic variances

AbstractIn this study, we introduce an innovative methodology aimed at enhancing Fisher's Linear Discriminant Analysis (LDA) in the context of high‐dimensional data classification scenarios, specifically addressing situations where each feature exhibits distinct variances. Our approach leverages Nonparametric Maximum Likelihood Estimation (NPMLE) techniques to estimate both the mean and variance parameters. By accommodating varying variances among features, our proposed method leads to notable improvements in classification performance. In particular, unlike numerous prior studies that assume the distribution of heterogeneous variances follows a right‐skewed inverse gamma distribution, our proposed method demonstrates excellent performance even when the distribution of heterogeneous variances takes on left‐skewed, symmetric, or right‐skewed forms. We conducted a series of rigorous experiments to empirically validate the effectiveness of our approach. The results of these experiments demonstrate that our proposed methodology excels in accurately classifying high‐dimensional data characterized by heterogeneous variances.

Integration of Mathematical Operators Based on Decision Boundary Complexity and Combinatorial Optimization for Improved Deep Learning Classifiers

The proliferation of complex diseases in livestock, such as lumpy skin disease, demands advanced diagnostic tools that can accurately classify and predict outbreaks. This study explores the integration of complex mathematical operators within deep learning classifiers to enhance their accuracy and efficiency in diagnosing lumpy skin disease. By focusing on the decision boundary complexity, which delineates different disease states in high-dimensional spaces, and employing combinatorial optimization techniques, we develop a novel framework that significantly improves classification performance. The methodology hinges on optimizing the configuration and combination of mathematical operators, such as gradient operators and higher-order derivatives, to refine feature extraction processes. This approach allows for a more nuanced understanding of the disease features that are critical for accurate classification. Using a dataset comprised of clinical and image data from infected cattle, our enhanced classifiers demonstrate a marked improvement in predictive accuracy compared to traditional deep learning models. The case study not only underscores the potential of integrating advanced mathematical concepts into deep learning but also sets a precedent for tackling similar challenges in veterinary medicine and beyond.

A deep belief rule base-based fault diagnosis method for complex systems

Complex systems are prone to faults due to their intricate structures, potentially impacting system stability. Therefore, fault diagnosis has become crucial for maintaining stable operation. In the field of complex systems, the combinatorial explosion problem in belief rule base (BRB) has attracted significant attention. The interdependence among system components leads to numerous variables and the need for rules, heightening model complexity. Regarding the combinatorial explosion problem, an improved belief rule network structure called deep BRB (DBRB) is proposed. First, the extreme gradient boosting (XGBoost) feature selection method is employed to choose the relatively important feature subset. Next, driven by the importance of features, different levels of features are input into the model, forming a complete and progressive network structure. Finally, the model undergoes the reasoning and optimization process. The effectiveness of the model is confirmed with a bearing fault dataset. After a comprehensive evaluation of multiple indicators, this method demonstrates a consistent improvement in classification performance as the depth increased. Moreover, compared to the traditional BRB model, this method notably reduces the number of parameters, improving its efficiency of processing complex data. In short, this method effectively tackles combinatorial explosion while ensuring model performance. The selection and assignment of feature subsets enhance the logic and readability of the model. Through the network structure, various fault features are captured well. This fault diagnosis method, rooted in the DBRB, offers a novel perspective on diagnosing complex system faults.

Rice phenology monitoring via ensemble classification for an extremely imbalanced multiclass dataset of hybrid remote sensing

This research provides a new approach for monitoring crops, especially rice planting, using hybrid remote sensing by aligning direct observations via Area Sampling Frame (ASF) surveys and earth observations via the LANDSAT 8 spectral indices. ASF surveys classify the land cover into eight groups, four associated with rice phenology. There's probably an imbalance between the classes in the multi-class ASF dataset. Often, when dealing with such extremely multiclass datasets, the performance of conventional classifiers is not adequate. Therefore, this study adopts a resampling technique to improve ensemble classification performance for handling extremely imbalanced multiclass ASF datasets. This research has added resampling methods, ROS (Random Over Sampling) and SCUT (SMOTE and Cluster-based Undersampling Technique), to the ensemble model, namely Bagging and Random Forest (RF), which have an impact on improving classification performance. The results showed that ROS-RF performed well in classifying multiclass on the ASF Survey with accuracy (90%), average sensitivity (90%), average specificity (99%), and balanced accuracy (94%). The variable importance (VI) with the highest ranking based on the Mean Decrease Gini (MDG) is the Modified Normalized Difference Water Index, which was recorded by the LANDSAT 8 satellite in one period before (MDWIt-1). However, the nine variables included in the classification generally contribute almost the same level of importance. The recommended classification strategy for highly imbalanced multi-class datasets is ROS-RF, which contributes to improving RF classification performance. The initial step was repeating minority samples until the class distribution was balanced before executing the classification stage.

Model-Based Approaches to Investigating Mismatch Responses in Schizophrenia.

Alterations of mismatch responses (ie, neural activity evoked by unexpected stimuli) are often considered a potential biomarker of schizophrenia. Going beyond establishing the type of observed alterations found in diagnosed patients and related cohorts, computational methods can yield valuable insights into the underlying disruptions of neural mechanisms and cognitive function. Here, we adopt a typology of model-based approaches from computational cognitive neuroscience, providing an overview of the study of mismatch responses and their alterations in schizophrenia from four complementary perspectives: (a) connectivity models, (b) decoding models, (c) neural network models, and (d) cognitive models. Connectivity models aim at inferring the effective connectivity patterns between brain regions that may underlie mismatch responses measured at the sensor level. Decoding models use multivariate spatiotemporal mismatch response patterns to infer the type of sensory violations or to classify participants based on their diagnosis. Neural network models such as deep convolutional neural networks can be used for improved classification performance as well as for a systematic study of various aspects of empirical data. Finally, cognitive models quantify mismatch responses in terms of signaling and updating perceptual predictions over time. In addition to describing the available methodology and reviewing the results of recent computational psychiatry studies, we offer suggestions for future work applying model-based techniques to advance the study of mismatch responses in schizophrenia.

Combining EEG Features and Convolutional Autoencoder for Neonatal Seizure Detection.

Neonatal epilepsy is a common emergency phenomenon in neonatal intensive care units (NICUs), which requires timely attention, early identification, and treatment. Traditional detection methods mostly use supervised learning with enormous labeled data. Hence, this study offers a semi-supervised hybrid architecture for detecting seizures, which combines the extracted electroencephalogram (EEG) feature dataset and convolutional autoencoder, called Fd-CAE. First, various features in the time domain and entropy domain are extracted to characterize the EEG signal, which helps distinguish epileptic seizures subsequently. Then, the unlabeled EEG features are fed into the convolutional autoencoder (CAE) for training, which effectively represents EEG features by optimizing the loss between the input and output features. This unsupervised feature learning process can better combine and optimize EEG features from unlabeled data. After that, the pre-trained encoder part of the model is used for further feature learning of labeled data to obtain its low-dimensional feature representation and achieve classification. This model is performed on the neonatal EEG dataset collected at the University of Helsinki Hospital, which has a high discriminative ability to detect seizures, with an accuracy of 92.34%, precision of 93.61%, recall rate of 98.74%, and F1-score of 95.77%, respectively. The results show that unsupervised learning by CAE is beneficial to the characterization of EEG signals, and the proposed Fd-CAE method significantly improves classification performance.

Stellar Classification with Vision Transformer and SDSS Photometric Images

With the development of large-scale sky surveys, an increasing number of stellar photometric images have been obtained. However, most stars lack spectroscopic data, which hinders stellar classification. Vision Transformer (ViT) has shown superior performance in image classification tasks compared to most convolutional neural networks (CNNs). In this study, we propose an stellar classification network based on the Transformer architecture, named stellar-ViT, aiming to efficiently and accurately classify the spectral class for stars when provided with photometric images. By utilizing RGB images synthesized from photometric data provided by the Sloan Digital Sky Survey (SDSS), our model can distinguish the seven main stellar categories: O, B, A, F, G, K, and M. Particularly, our stellar-ViT-gri model, which reaches an accuracy of 0.839, outperforms traditional CNNs and the current state-of-the-art stellar classification network SCNet when processing RGB images synthesized from the gri bands. Furthermore, with the introduction of urz band data, the overall accuracy of the stellar-ViT model reaches 0.863, further demonstrating the importance of additional band information in improving classification performance. Our approach showcases the effectiveness and feasibility of using photometric images and Transformers for stellar classification through simple data augmentation strategies and robustness analysis of training dataset sizes. The stellar-ViT model maintains good performance even in small sample scenarios, and the inclusion of urz band data reduces the likelihood of misclassifying samples as lower-temperature subtypes.

Breast cancer diagnosis using support vector machine optimized by improved quantum inspired grey wolf optimization

A prompt diagnosis of breast cancer in its earliest phases is necessary for effective treatment. While Computer-Aided Diagnosis systems play a crucial role in automated mammography image processing, interpretation, grading, and early detection of breast cancer, existing approaches face limitations in achieving optimal accuracy. This study addresses these limitations by hybridizing the improved quantum-inspired binary Grey Wolf Optimizer with the Support Vector Machines Radial Basis Function Kernel. This hybrid approach aims to enhance the accuracy of breast cancer classification by determining the optimal Support Vector Machine parameters. The motivation for this hybridization lies in the need for improved classification performance compared to existing optimizers such as Particle Swarm Optimization and Genetic Algorithm. Evaluate the efficacy of the proposed IQI-BGWO-SVM approach on the MIAS dataset, considering various metric parameters, including accuracy, sensitivity, and specificity. Furthermore, the application of IQI-BGWO-SVM for feature selection will be explored, and the results will be compared. Experimental findings demonstrate that the suggested IQI-BGWO-SVM technique outperforms state-of-the-art classification methods on the MIAS dataset, with a resulting mean accuracy, sensitivity, and specificity of 99.25%, 98.96%, and 100%, respectively, using a tenfold cross-validation datasets partition.

Predictive Models for Suicide Attempts in Major Depressive Disorder and the Contribution of EPHX2: A Pilot Integrative Machine Learning Study

Suicide is a major public health problem caused by a complex interaction of various factors. Major depressive disorder (MDD) is the most prevalent psychiatric disorder associated with suicide; therefore, it is essential to prioritize suicide prediction and prevention within this population. Integrated information from different dimensions, including personality, cognitive function, and social and genetic factors, is necessary to improve the performance of predictive models. Besides, recent studies have indicated the critical roles for EPHX2/P2X2 in the pathophysiology of MDD. Our previous studies found an association of EPHX2 and P2X2 with suicide in MDD. This study is aimed at (1) establishing predictive models with integrated information to distinguish MDD from healthy volunteers, (2) estimating the suicide risk of MDD, and (3) determining the contribution of EPHX2/P2X2. This cross-sectional study was conducted on 472 prospectively collected participants. The machine learning (ML) technique using Extreme Gradient Boosting (XGBoost) classifier was employed to evaluate the performance and relative importance of the extracted characteristics in recognising patients with MDD and depressed suicide attempters (DSA). In independent validation set, the model with clinical and cognitive information could recognise MDD with an area under the receiver operating characteristic curve (AUC) of 0.938 (95% confidence interval (CI), 0.898–0.977), and genetic information did not improve classification performance. The model with clinical, cognitive, and genetic information resulted in a significantly higher AUC of 0.801 (95% CI, 0.719–0.884) for identifying DSA than the model with only clinical information, in which the three single nucleotide polymorphisms of EPHX2 showed important roles. This study successfully established step-by-step predictive ML models to estimate the risk of suicide attempts in MDD. We found that EPHX2 can help improve the performance of suicidal predictive models. This trial is registered with NCT05575713.

A MeanShift-guided oversampling with self-adaptive sizes for imbalanced data classification

The imbalanced data classification has gained popularity in machine learning research domain due to its prevalence in numerous applications and its difficulty. However, the majority of contemporary work primarily focuses on addressing between-class imbalance issues. Previous researches have shown that combined with other elements, such as within-class imbalance, small sample size and the presence of small disjuncts, the imbalanced data significantly increase the difficulties for the traditional classifiers to learn. Therefore, we propose a novel MeanShift-guided oversampling with self-adaptive sizes for imbalanced data classification. The proposed MeanShift-guided oversampling technique can simultaneously consider the distribution of minority class and majority class within the sphere with the current minority instance as its center, which can favor addressing small sample size and avoiding overlapping issues often caused by the nearest neighbor (NN)-based oversampling techniques. The incorporation of random vector and flexible cut-off mechanism for vector length can enhance the diversity among the generated synthetic minority instances and avoid overlapping, which makes it suitable for small sample size and small disjuncts problems. To address between-class and within-class imbalance issues, we also introduce a self-adaptive sizes assignment strategy for each minority instance to be oversampled, where the assigned size is inversely proportional to its density and its distance from the majority class. In addition to eliminating within-class imbalance, the strategy can ensure that the informative border minority instances have more opportunities to be oversampled, thus improving classification performance. Extensive experimental results on some datasets with different distributions and imbalance ratios show the proposed algorithm outperforms other compared ones with significant difference.

Joint negative–positive-learning based sample reweighting for hyperspectral image classification with label noise

Deep neural networks (DNNs) have been widely used for hyperspectral image (HSI) classification. However, the superior performance of DNNs requires accurately labeled samples. In various applications, label noise in large-scale HSI data is often unavoidable and adversely affects the performance of a classifier. Therefore, establishing a robust classification model from an existing DNN classifier for the HSI dataset with label noise is a significant and challenging problem. In this report, we propose a meta-learning reweighting framework based on a joint positive and negative learning (JPNL) method which can adaptively reweight labeled samples to improve the robustness of the classification model. Experimental results on two commonly used hyperspectral imaging (HSI) datasets contaminated with label noise demonstrate significant improvements in classification performance and robustness achieved by the proposed framework, as compared to existing classifiers.