Semi-supervised Convolutional Neural Network Research Articles

Extracting fetal heart signals from Doppler using semi-supervised convolutional neural networks.

Cardiotocography (CTG) measurements are critical for assessing fetal wellbeing during monitoring, and accurate assessment requires well-traceable CTG signals. The current FHR calculation algorithm, based on autocorrelation to Doppler ultrasound (DUS) signals, often results in periods of loss owing to its inability to differentiate signals. We hypothesized that classifying DUS signals by type could be a solution and proposed that an artificial intelligence (AI)-based approach could be used for classification. However, limited studies have incorporated the use of AI for DUS signals because of the limited data availability. Therefore, this study focused on evaluating the effectiveness of semi-supervised learning in enhancing classification accuracy, even in limited datasets, for DUS signals. Data comprising fetal heartbeat, artifacts, and two other categories were created from non-stress tests and labor DUS signals. With labeled and unlabeled data totaling 9,600 and 48,000 data points, respectively, the semi-supervised learning model consistently outperformed the supervised learning model, achieving an average classification accuracy of 80.9%. The preliminary findings indicate that applying semi-supervised learning to the development of AI models using DUS signals can achieve high generalization accuracy and reduce the effort. This approach may enhance the quality of fetal monitoring.

SscNOVA: a semi-supervised convolutional neural network for predicting functional regulatory variants in autoimmune diseases.

Genome-wide association studies (GWAS) have identified thousands of variants in the human genome with autoimmune diseases. However, identifying functional regulatory variants associated with autoimmune diseases remains challenging, largely because of insufficient experimental validation data. We adopt the concept of semi-supervised learning by combining labeled and unlabeled data to develop a deep learning-based algorithm framework, sscNOVA, to predict functional regulatory variants in autoimmune diseases and analyze the functional characteristics of these regulatory variants. Compared to traditional supervised learning methods, our approach leverages more variants' data to explore the relationship between functional regulatory variants and autoimmune diseases. Based on the experimentally curated testing dataset and evaluation metrics, we find that sscNOVA outperforms other state-of-the-art methods. Furthermore, we illustrate that sscNOVA can help to improve the prioritization of functional regulatory variants from lead single-nucleotide polymorphisms and the proxy variants in autoimmune GWAS data.

Genetic algorithm-based semisupervised convolutional neural network for real-time monitoring of Escherichia coli fermentation of recombinant protein production using a Raman sensor.

As a non-destructive sensing technique, Raman spectroscopy is often combined with regression models for real-time detection of key components in microbial cultivation processes. However, achieving accurate model predictions often requires a large amount of offline measurement data for training, which is both time-consuming and labor-intensive. In order to overcome the limitations of traditional models that rely on large datasets and complex spectral preprocessing, in addition to the difficulty of training models with limited samples, we have explored a genetic algorithm-based semi-supervised convolutional neural network (GA-SCNN). GA-SCNN integrates unsupervised process spectral labeling, feature extraction, regression prediction, and transfer learning. Using only an extremely small number of offline samples of the target protein, this framework can accurately predict protein concentration, which represents a significant challenge for other models. The effectiveness of the framework has been validated in a system of Escherichia coli expressing recombinant ProA5M protein. By utilizing the labeling technique of this framework, the available dataset for glucose, lactate, ammonium ions, and optical density at 600 nm (OD600) has been expanded from 52 samples to 1302 samples. Furthermore, by introducing a small component of offline detection data for recombinant proteins into the OD600 model through transfer learning, a model for target protein detection has been retrained, providing a new direction for the development of associated models. Comparative analysis with traditional algorithms demonstrates that the GA-SCNN framework exhibits good adaptability when there is no complex spectral preprocessing. Cross-validation results confirm the robustness and high accuracy of the framework, with the predicted values of the model highly consistent with the offline measurement results.

Melanoma Breslow Thickness Classification using Ensemble-based Knowledge Distillation with Semi-supervised Convolutional Neural Networks.

Melanoma is considered a global public health challenge and is responsible for more than 90% deaths related to skin cancer. Although the diagnosis of early melanoma is the main goal of dermoscopy, the discrimination between dermoscopic images of in situ and invasive melanomas can be a difficult task even for experienced dermatologists. Recent advances in artificial intelligence in the field of medical image analysis show that its application to dermoscopy with the aim of supporting and providing a second opinion to the medical expert could be of great interest. In this work, four datasets from different sources were used to train and evaluate deep learning models on in situ versus invasive melanoma classification and on Breslow thickness prediction. Supervised learning and semi-supervised learning using a multi-teacher ensemble knowledge distillation approach were considered and evaluated using a stratified 5-fold cross-validation scheme. The best models achieved AUCs of 0.6186 ±0.0410 and of 0.7501 ±0.0674 on the former and latter classification tasks, respectively. The best results were obtained using semi-supervised learning, with the best model achieving 0.7751 and 0.8164 AUC, respectively. The results obtained show that semi-supervised learning could improve the performance of trained models in different melanoma classification tasks compared to supervised learning. Automatic deep learning-based diagnosis systems could support medical professionals in their decision, serving as a second opinion or as a triage tool for medical centers.

Automatic Classification of White Blood Cells Using a Semi-Supervised Convolutional Neural Network

Automatic Classification of White Blood Cells Using a Semi-Supervised Convolutional Neural Network

ScribbleDom: Using Scribble-Annotated Histology Images to Identify Domains in Spatial Transcriptomics Data.

Spatial domain identification is a very important problem in the field of Spatial Transcriptomics (ST). The state-of-the-art solutions to this problem focus on unsupervised methods, as there is lack of data for a supervised learning formulation. The results obtained from these methods highlight significant opportunities for improvement. In this paper, we propose a potential avenue for enhancement through the development of a semi-supervised convolutional neural network (CNN) based approach. Named ScribbleDom, our method leverages human expert's input as a form of semi-supervision, thereby seamlessly combines the cognitive abilities of human experts with the computational power of machines. ScribbleDom incorporates a loss function that integrates two crucial components: similarity in gene expression profiles and adherence to the valuable input of a human annotator through scribbles on histology images, providing prior knowledge about spot labels. The spatial continuity of the tissue domains is taken into account by extracting information on the spot micro-environment through convolution filters of varying sizes, in the form of Inception blocks. By leveraging this semi-supervised approach, ScribbleDom significantly improves the quality of spatial domains, yielding superior results both quantitatively and qualitatively. Our experiments on several benchmark datasets demonstrate the clear edge of ScribbleDom over state-of-the-art methods-between 1.82% to 169.38% improvements in Adjusted Rand Index (ARI) for 9 of the 12 Human DLPFC samples, and 15.54% improvement in the Melanoma cancer dataset. Notably, when the expert input is absent, ScribbleDom can still operate, in a fully unsupervised manner like the state-of-the-art methods, and produces results that remain competitive. Source code is available at Github (https://github.com/1alnoman/ScribbleDom) and Zenodo (https://zenodo.org/badge/latestdoi/681572669). Supplementary data are available at Bioinformatics online.

Natural language processing to convert unstructured COVID-19 chest-CT reports into structured reports

BackgroundStructured reporting has been demonstrated to increase report completeness and to reduce error rate, also enabling data mining of radiological reports. Still, structured reporting is perceived by radiologists as a fragmented reporting style, limiting their freedom of expression. PurposeA deep learning-based natural language processing method was developed to automatically convert unstructured COVID-19 chest CT reports into structured reports. MethodsTwo hundred-two COVID-19 chest CT were retrospectively reviewed by two experienced radiologists, who wrote for each exam a free-form text radiological report and coherently filled the template provided by the Italian Society of Medical and Interventional Radiology, used as ground-truth. A semi-supervised convolutional neural network was implemented to extract 62 categorical variables from the report. Two iterations were carried-out, the first without fine-tuning, the second one performing a fine-tuning. The performance was measured using the mean accuracy and the F1 mean score. An error analysis was performed to identify errors entirely attributable to incorrect processing of the model. ResultsThe algorithm achieved a mean accuracy of 93.7% and an F1 score 93.8% in the first iteration. Most of the errors were exclusively attributable to wrong inference (46%). In the second iteration the model achieved for both parameters 95,8% and percentage of errors attributable to wrong inference decreased to 26%. ConclusionsThe convolutional neural network achieved an optimal performance in the automated conversion of free-form text into structured radiological reports, overcoming all the limitation attributed to structured reporting and finally paving the way for data mining of radiological report.

Application of semi-supervised convolutional neural network regression model based on data augmentation and process spectral labeling in Raman predictive modeling of cell culture processes

Raman spectroscopy coupled to chemometric model-based technology have a great potential for real-time monitoring of critical process parameters in animal cell culture. The measurement of off-line reference values required for calibration of monitoring procedures is complex, time- and resource-intensive, whereas the acquisition of spectra is a seconds- or minutes-based process. Usually, less than 5% spectra is labeled with off-line data during calibration steps, thus a large amount of spectra is wasted. In this study, we proposed a Semi-Supervised Convolutional Neural Network Regression (SSCNNR) model framework for Raman model establishment based on data augmentation and process spectra labeling. Compared with the original CNNR model, the size of the calibration dataset was expanded from 132 samples to nearly 8500, and the constructed SSCNNR models obtained a significantly improved accuracy in predicting glucose, glutamine, asparagine, and ammonium, with RMSEP values decreased by 29.1%, 37.3%, 38.3% and 7%, respectively. Compared with traditional modeling approach of PLSR and SVR, the SSCNNR models were proved to accurately predict the trend of substance concentration during frequent feeding process. Based on this framework, potential efforts included improving model prediction accuracy by hyperparameters optimization and accommodating scenarios of culture change by transfer learning.

Aesthetic normalization of gender in the Instagram application: A portrait of the Brazilian woman

Social media’s embedded artificial intelligence (AI) plays a role as a behavior shaping in modern society. Two crucial aspects of this tool are related to content production and reproduction that allow exceptional dissemination capacity while creating a feeling of engagement among users. This research aims to assess the aesthetic normalization of brazilian women as portrayed on Instagram's top-ranked content. Data analysis follows a two-step methodological approach that searches and stores public profile’s meta-images as displayed on Instagram and, then, automatically processes gender labeling and features classification according to pre-selected aesthetic standards commonly associated with Brazilian women. A human auditing process was set to increase accuracy in the sample categorization. Methodology development uses état de l'art developments of computer visions and a semi-supervised Convolutional Neural Networks (CNN) machine learning model. The typical woman is portrayed in full-body, dressed in swimming suits or gym-outfit, and aged around 30 years old. Altogether, the results suggest the recurrence of real-world aesthetic-related stereotypes for Brazilian women on Instagram.

An Intelligent Fault Diagnosis Based on Adversarial Generating Module and Semi-supervised Convolutional Neural Network.

Aiming at the existing problems in machinery monitoring data such as high cost of labeling and lack of typical failure samples, this paper launches a research on the semi-supervised-style intelligent fault diagnosis. Taking a great mount of unlabeled data and only a small quantity of labeled data as inputs, a novel fault diagnosis framework based on adversarial generating module and semi-supervised convolutional neural network (SSCNN) is proposed. Firstly, a semi-supervised learning module based on manifold-regularization-based fuzzy clustering discrimination (MRFCD) is proposed to make full use of the valuable fault-related information contained in unlabeled data. Secondly, MRFCD was introduced into CNN to construct pseudo-labels and estimate the objective function of unlabeled data. Then, the semi-supervised deep-learning-module-based MRFCD-SSCNN is established. Thirdly, to enhance the effect of MRFCD-SSCNN, generative adversarial network (GAN) was utilized to increase the size of training data under failure conditions. The framework based on GAN-MRFCD-SSCNN is proposed to achieve semi-supervised style intelligent fault diagnosis. To verify the performance of the diagnostic framework, vibrational signals of main reducer collected from actual test rig are employed. The comparative results confirm that the proposed framework outperforms some classical semi-supervised diagnostic models and achieves the accuracy of 96.2% using only 400 labeled samples.

Proposing Novel Data Analytics Method for Anatomical Landmark Identification from Endoscopic Video Frames.

Background The anatomical landmarks contain the characteristics that are used to guide the gastroenterologists during the endoscopy. The expert can also ensure the completion of examination with the help of the anatomical landmarks. Automatic detection of anatomical landmarks in endoscopic video frames can be helpful for guiding the physicians during screening the gastrointestinal tract (GI). Method This study presents an automatic novel method for anatomical landmark detection of GI tract from endoscopic video frames based on semisupervised deep convolutional neural network (CNN) and compares the results with supervised CNN model. We consider the anatomical landmarks from Kvasir dataset that includes 500 images for each class of Z-line, pylorus, and cecum. The resolution of these images varies from 750 × 576 up to 1920 × 1072 pixels. Result Experimental results show that the supervised CNN has highly desirable performance with accuracy of 100%. Also, our proposed semisupervised CNN can compete with a slight difference similar to the CNN model. Our proposed semisupervised model trained using 1, 5, 10, and 20 percent of training data records as labeled training dataset has the average accuracy of 83%, 98%, 99%, and 99%, respectively. Conclusion The main advantage of our proposed method is achieving the high accuracy with small amount of labeled data without spending time for labeling more data. The strength of our proposed method saves the required labor, cost, and time for data labeling.

SC-EADNet: A Self-Supervised Contrastive Efficient Asymmetric Dilated Network for Hyperspectral Image Classification

Unsupervised and semisupervised feature learning has recently emerged as an effective way to reduce the reliance on expensive data collection and annotation for hyperspectral image (HSI) analysis. Existing unsupervised and semisupervised convolutional neural network (CNN)-based HSI classification works still face two challenges: underutilization of pixel-wise multiscale contextual information for feature learning and expensive computational cost, for example, large floating-point operations per seconds (FLOPs), due to the lack of lightweight design. To utilize the unlabeled pixels in the HSIs more efficiently, we propose a self-supervised contrastive efficient asymmetric dilated network (SC-EADNet) for HSI classification. There are two novelties in the SC-EADNet. First, a self-supervised multiscale pixel-wise contextual feature learning model is proposed, which generates multiple patches around each hyperspectral pixel and develops a contrastive learning framework to learn from these patches for HSI classification. Second, a lightweight feature extraction network EADNet, composed of multiple plug-and-play efficient asymmetric dilated convolution (EADC) blocks, is designed and inserted into the contrastive learning framework. The EADC block adopts different dilation rates to capture the spatial information of objects with varying shapes and sizes. Compared with other unsupervised, semisupervised, and supervised learning methods, our SC-EADNet provides competitive classification performance on four hyperspectral datasets, including Indian Pines, Pavia University, Salinas, and Houston 2013, but few FLOPs and fast computational speed.

Deep Reinforcement Learning for Semisupervised Hyperspectral Band Selection

Band selection is an important step in efficient processing of hyperspectral images (HSIs), which can be seen as the combination of powerful band search technique and effective evaluation criterion. The existing deep-learning-based methods make the network parameters sparse to search the spectral bands using threshold-based functions or regularization terms. These methods may lead to an intractable optimization problem. Furthermore, these methods need to repeatedly train deep networks for evaluating candidate band subsets. In this article, we formalize hyperspectral band selection as a reinforcement learning (RL) problem. Band search is regarded as a sequential decision-making process, where each state in the search space is a feasible band subset. To evaluate each state, a semisupervised convolutional neural network (CNN), called EvaluateNet, is constructed by adding the intraclass compactness constraint of both limited labeled and sufficient unlabeled samples. A simple stochastic band sampling method is designed to train EvaluateNet, making it possible to efficiently evaluate without any fine-tuning. In RL, new reward functions are defined by taking the EvaluateNet and the penalty of repeated selection into account. Finally, advantage actor–critic algorithms are designed to explore in the state space and select the band subset according to the expected accumulated reward. The experimental results on HSI data sets demonstrate the effectiveness and efficiency of the proposed algorithms for hyperspectral band selection.

Semi-Active Convolutional Neural Networks for Hyperspectral Image Classification

Owing to the powerful data representation ability of deep learning (DL) techniques, tremendous progress has been recently made in hyperspectral image (HSI) classification. Convolutional neural network (CNN), as a main part of the DL family, has been proven to be considerably effective to extract spatial-spectral features for HSIs. Nevertheless, its classification performance, to a great extent, depends on the quality and quantity of samples in the network training process. To select those samples, either labeled or unlabeled, that can be used to enhance the generalization ability of CNNs and further improve the classification accuracy, we propose an iterative semi-supervised CNNs framework by means of active learning and superpixel segmentation techniques, dubbed as semi-active CNNs (SA-CNNs), for HSI classification. More specifically, we start to pre-train a CNNs-based model on a small-scale unbiased labeled set and infer unlabeled data using the trained model, i.e., generating pseudo-labels. Then, the reliable samples, which consist of two parts: high label-homogeneity and most informativeness, are actively selected from superpixel segments. These selected labeled and unlabeled samples with their labels and pseudo-labels are re-fed into the next-round network training. Moreover, three different schedules, i.e., <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">log</i> -, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">exp</i> -, and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">linear</i> -schedules, are progressively adopted to fully explore their potentials in sample selection, until a labeling budget is finally reached. Extensive experiments are conducted on three benchmark HSI datasets, demonstrating substantial performance improvements of the proposed SA-CNNs over other similar competitors.

Online Capacity Estimation for Lithium-Ion Batteries Based on Semi-Supervised Convolutional Neural Network

Accurate capacity estimation can ensure the safe and reliable operation of lithium-ion batteries in practical applications. Recently, deep learning-based capacity estimation methods have demonstrated impressive advances. However, such methods suffer from limited labeled data for training, i.e., the capacity ground-truth of lithium-ion batteries. A capacity estimation method is proposed based on a semi-supervised convolutional neural network (SS-CNN). This method can automatically extract features from battery partial-charge information for capacity estimation. Furthermore, a semi-supervised training strategy is developed to take advantage of the extra unlabeled sample, which can improve the generalization of the model and the accuracy of capacity estimation even in the presence of limited labeled data. Compared with artificial neural networks and convolutional neural networks, the proposed method is demonstrated to improve capacity estimation accuracy.

Semi-Supervised Convolutional Long Short-Term Memory Neural Networks for Time Series Land Cover Classification

Time series images with temporal features are beneficial to improve the classification accuracy. For abstract temporal and spatial contextual information, deep neural networks have become an effective method. However, there is usually a lack of sufficient samples in network training: one is the loss of images or the discontinuous distribution of time series data because of the inevitable cloud cover, and the other is the lack of known labeled data. In this paper, we proposed a Semi-supervised convolutional Long Short-Term Memory neural network (SemiLSTM) for time series remote sensing images, which was validated on three data sets with different time distributions. It achieves an accurate and automated land cover classification via a small number of labeled samples and a large number of unlabeled samples. Besides, it is a robust classification algorithm for time series optical images with cloud coverage, which reduces the requirements for cloudless remote sensing images and can be widely used in areas that are often obscured by clouds, such as subtropical areas. In conclusion, this method makes full advantage of spectral-spatial-temporal characteristics under the condition of limited training samples, especially expanding time context information to enhance classification accuracy.

Analysis of unlabeled lung sound samples using semi-supervised convolutional neural networks

Lung sounds convey valuable information relevant to human respiratory health. Therefore, it is important to classify lung sounds for early diagnoses of respiratory disorders. In recent years, computerized lung sound analysis with machine learning algorithms has attracted researchers, especially the state-of-the-art convolutional neural network (CNN). However, most of these algorithms require a large number of labeled respiratory sound samples, which is time- and cost-consuming. Based on a four-layers CNN, this study proposes graph semi-supervised CNNs (GS-CNNs), which can classify respiratory sounds into normal, crackle and wheeze ones with only a small labeled sample size and a large unlabeled sample size. The graph of respiratory sounds (Graph-RS) with labeled and unlabeled respiratory sound samples as vertexes is first constructed, which can indicate not only the reasonable metric information but also the relationship of all the samples. Then, GS-CNNs are developed by adding the information extracted from Graph-RS to the loss function of the original CNN. The added information enables the GS-CNNs to regulate the structure of the original CNN, thus enhancing classification accuracy. The GS-CNNs are evaluated by experiments with the samples collected by electronic stethoscope. Results demonstrate that the proposed GS-CNNs outperform the original CNN, and that the more information from Graph-RS is used, the better recognition effect will be achieved.

Intelligent fault diagnosis of rolling bearings using a semi-supervised convolutional neural network

The success of convolutional neural networks (CNNs) in intelligent fault diagnosis is largely dependent on massive amounts of labelled data. In a real-world case, however, massive amounts of labelled data are difficult or costly to collect, whereas abundant unlabelled data are often available. To utilize such unlabelled data, a novel method using a semi-supervised convolutional neural network (SSCNN) for intelligent fault diagnosis of bearings is proposed. First, a 1-d CNN is applied to learn class space features and generate class probabilities of unlabelled samples, based on which a class probability maximum margin criterion (CPMMC) method is used to construct the loss function of unlabelled samples. Then, the constructed loss function, which aims to maximise the inter-class distance of class space features and minimise the intra-class distance of class space features, is integrated into the cross-entropy loss function of the CNN, and the SSCNN is established. Finally, the SSCNN model is applied to analyse the vibration signals collected from rolling bearings, and a novel intelligent fault diagnosis method using the SSCNN is proposed. Two datasets are employed to validate the effectiveness of the proposed methodology. The results show that the established SSCNN can effectively utilise unlabelled samples to train the model and enhance its fault diagnosis performance. Through a comparison with commonly used semi-supervised deep learning methods, the superiority of the proposed method is validated.

TIME SERIES LAND COVER CLASSIFICATION BASED ON SEMI-SUPERVISED CONVOLUTIONAL LONG SHORT-TERM MEMORY NEURAL NETWORKS

Abstract. Time series imagery containing high-dimensional temporal features are conducive to improving classification accuracy. With the plenty accumulation of historical images, the inclusion of time series data becomes available to utilize, but it is difficult to avoid missing values caused by cloud cover. Meanwhile, seeking a large amount of training labels for long time series also makes data collection troublesome. In this study, we proposed a semi-supervised convolutional long short-term memory neural network (Semi-LSTM) in long time series which achieves an accurate and automated land cover classification with a small proportion of labels. Three main contributions of this work are summarized as follows: i) the proposed method achieve an excellent classification via a small group of labels in long time series data, and reducing dependence of training labels; ii) it is a robust algorithm in accuracy for the influence of noise, and reduces the requirements of sequential data for cloudless and lossless images; and iii) it makes full advantage of spectral-spatial-temporal features, especially expanding time context information to enhance classification accuracy. Finally, the proposed network is validated on time series imagery from Landsat 8. All quantitative analyses and evaluation indicators of the experimental results demonstrate competitive performance in the suggested modes.

A semi-supervised convolutional neural network based on subspace representation for image classification

This work presents a shallow network based on subspaces with applications in image classification. Recently, shallow networks based on PCA filter banks have been employed to solve many computer vision-related problems including texture classification, face recognition, and scene understanding. These approaches are robust, with a straightforward implementation that enables fast prototyping of practical applications. However, these architectures employ either unsupervised or supervised learning. As a result, they may not achieve highly discriminative features in more complicated computer vision problems containing variations in camera motion, object’s appearance, pose, scale, and texture, due to drawbacks related to each learning paradigm. To cope with this disadvantage, we propose a semi-supervised shallow network equipped with both unsupervised and supervised filter banks, presenting representative and discriminative abilities. Besides, the introduced architecture is flexible, performing favorably on different applications whose amount of supervised data is an issue, making it an attractive choice in practice. The proposed network is evaluated on five datasets. The results show improvement in terms of prediction rate, comparing to current shallow networks.