Deep Network Classifier Research Articles

Development and evaluation of a deep neural network for histologic classification of renal cell carcinoma on biopsy and surgical resection slides

Renal cell carcinoma (RCC) is the most common renal cancer in adults. The histopathologic classification of RCC is essential for diagnosis, prognosis, and management of patients. Reorganization and classification of complex histologic patterns of RCC on biopsy and surgical resection slides under a microscope remains a heavily specialized, error-prone, and time-consuming task for pathologists. In this study, we developed a deep neural network model that can accurately classify digitized surgical resection slides and biopsy slides into five related classes: clear cell RCC, papillary RCC, chromophobe RCC, renal oncocytoma, and normal. In addition to the whole-slide classification pipeline, we visualized the identified indicative regions and features on slides for classification by reprocessing patch-level classification results to ensure the explainability of our diagnostic model. We evaluated our model on independent test sets of 78 surgical resection whole slides and 79 biopsy slides from our tertiary medical institution, and 917 surgical resection slides from The Cancer Genome Atlas (TCGA) database. The average area under the curve (AUC) of our classifier on the internal resection slides, internal biopsy slides, and external TCGA slides is 0.98 (95% confidence interval (CI): 0.97–1.00), 0.98 (95% CI: 0.96–1.00) and 0.97 (95% CI: 0.96–0.98), respectively. Our results suggest that the high generalizability of our approach across different data sources and specimen types. More importantly, our model has the potential to assist pathologists by (1) automatically pre-screening slides to reduce false-negative cases, (2) highlighting regions of importance on digitized slides to accelerate diagnosis, and (3) providing objective and accurate diagnosis as the second opinion.

An automated COVID-19 detection based on fused dynamic exemplar pyramid feature extraction and hybrid feature selection using deep learning

The new coronavirus disease known as COVID-19 is currently a pandemic that is spread out the whole world. Several methods have been presented to detect COVID-19 disease. Computer vision methods have been widely utilized to detect COVID-19 by using chest X-ray and computed tomography (CT) images. This work introduces a model for the automatic detection of COVID-19 using CT images. A novel handcrafted feature generation technique and a hybrid feature selector are used together to achieve better performance. The primary goal of the proposed framework is to achieve a higher classification accuracy than convolutional neural networks (CNN) using handcrafted features of the CT images. In the proposed framework, there are four fundamental phases, which are preprocessing, fused dynamic sized exemplars based pyramid feature generation, ReliefF, and iterative neighborhood component analysis based feature selection and deep neural network classifier. In the preprocessing phase, CT images are converted into 2D matrices and resized to 256 × 256 sized images. The proposed feature generation network uses dynamic-sized exemplars and pyramid structures together. Two basic feature generation functions are used to extract statistical and textural features. The selected most informative features are forwarded to artificial neural networks (ANN) and deep neural network (DNN) for classification. ANN and DNN models achieved 94.10% and 95.84% classification accuracies respectively. The proposed fused feature generator and iterative hybrid feature selector achieved the best success rate, according to the results obtained by using CT images.

Functional deep echo state network improved by a bi-level optimization approach for multivariate time series classification

The multivariate time series (MTS) classification is one of the major tasks of time series data mining. Many methods have been proposed to investigate the MTS classification. Among them, the method based on feature representation is the most popular and widely used one. However, there exist some shortcomings for this method, such as unsatisfactory accuracy, being sensitive to noise and not able to fully make use of time series data attributes. In order to overcome these disadvantages, we propose a new method called functional deep echo state network (FDESN) for MTS classification that utilizes two special operators: temporal aggregation and spatial aggregation. In general, the parameters of the FDESN are determined by random selection, human experience or trial and error. This may increase the complexity of the FDESN or reduce the accuracy of the FDESN. In this study, a novel bi-level optimization approach is proposed to optimize the parameters of the FDESN. The parameter selection problem in the FDESN is transformed into the bi-level optimization problem. The state transition algorithm (STA) is used to solve the bi-level optimization problem. Finally, the experimental results show that the proposed method is superior to other methods. In addition, the proposed method is successfully applied to anode condition identification in aluminum electrolysis. For the aluminum electrolysis datasets, the proposed method improved the average classification accuracy by about 3.5% compared with the other methods. For a specific aluminum electrolysis dataset ACS2504, the classification accuracy significantly increased from 77.92% to 82.69% by using the proposed method.

Deep Neural Network for Somatic Mutation Classification

The detection and characterization of somatic mutations have become the important means to analyze the occurrence and development of cancer and, ultimately, will help to select effective and precise treatment for specific cancer patients. It is very difficult to detect somatic mutations accurately from the massive sequencing data. In this paper, a forest-graph-embedded deep feed-forward network (forgeNet) is utilized to detect somatic mutations from the sequencing data. In forgeNet, the random forest (RF) or Gradient Boosting Machine (GBM) and graph-embedded deep feed-forward network (GEDFN) are utilized to extract features and implement classification, respectively. Three real somatic mutation datasets collected from 48 triple-negative breast cancers are utilized to test the somatic mutation detection performances of forgeNet. The detection results show that forgeNet could make the 0.05%–0.424% improvements in terms of area under the curve (AUC) compared with support vector machines and random forest.

Deep multi-level fusion network for multi-source image pixel-wise classification

For multi-source image pixel-wise classification, each image information is different and complementary in the same area or scene. However, how to integrate them for decision-making is a difficult problem. In this paper, we focus on the characteristics of multi-source image and propose a novel pixel-wise classification method, named deep multi-level fusion network. The proposed method is to classify multi-sensor data including very high-resolution (VHR) RGB imagery, hyperspectral imagery (HSI) and multispectral light detection and ranging (MS-LiDAR) point cloud data. First, a deep spectral–spatial attention network is proposed to process HSI and MS-LiDAR images and get a learned classification map, which is based on feature level fusion. Next, a down-superpixel segmentation algorithm is proposed to get a segmentation result for VHR RGB imagery. Finally, the feature level fusion results are refinement by the down-superpixel segmentation results on the decision level, and get the final result. Extensive experiments and analyses on the data set grss_dfc_2018 demonstrate that the proposed multi-level fusion network can achieve a better result in the multi-source image pixel-wise classification.

Fractional-atom search algorithm-based deep recurrent neural network for cancer classification

Deep learning has been paid great attention in several fields for the good reason, and their results achieved were not before possible. In deep learning, the models are trained based on huge labelled dataset and the neural network architectures consist of several layers. Still, the methods suffer to achieve expected higher performance and introduce poor classification accuracy. Towards improving the classification performance, this paper introduces the deep learning networks for the cancer classification application. The main aim is to develop the optimized model for classification application. Accordingly, the input data is pre-processed using Log transformation for converting the data to its uniform value range. Then, the feature selection is done based on Wrapper approach to select the important features for classifying the cancer. Once the features are selected, the cancer classification is performed using Deep Recurrent Neural Network (Deep RNN), which is trained by the proposed Fractional-Atom Search Algorithm (Fractional-ASO). The Fractional-ASO is designed by integrating Fractional Calculus with Atom Search Optimization (ASO). The performance of the Fractional-ASO-based Deep RNN is evaluated in terms of accuracy, True Positive Rate (TPR) and True Negative Rate (TNR). The proposed Fractional-ASO -based Deep RNN method achieves the maximal accuracy of 92.87%, maximal TPR of 92.87%, and the maximal TNR of 93.48% using colon dataset.

Ensemble diversified learning for image classification with noisy labels

In this work, we develop a new approach for learning a deep neural network for image classification with noisy labels using ensemble diversified learning. We first partition the training set into multiple subsets with diversified image characteristics. For each subset, we train a separate deep neural network image classifier. These networks are then used to encode the input image into different feature vectors, providing diversified observations of the input image. The encoded features are then fused together and further analyzed by a decision network to produce the final classification output. We study image classification on noisy labels with and without the access to clean samples. Our extensive experimental results on the CIFAR-10 and MNIST datasets demonstrate that our proposed method outperforms existing methods by a large margin.

A novel lightweight deep convolutional neural network for early detection of oral cancer.

To develop a lightweight deep convolutional neural network (CNN) for binary classification of oral lesions into benign and malignant or potentially malignant using standard real-time clinical images. A small deep CNN, that uses a pretrained EfficientNet-B0 as a lightweight transfer learning model, was proposed. A data set of 716 clinical images was used to train and test the proposed model. Accuracy, specificity, sensitivity, receiver operating characteristics (ROC) and area under curve (AUC) were used to evaluate performance. Bootstrapping with 120 repetitions was used to calculate arithmetic means and 95% confidence intervals (CIs). The proposed CNN model achieved an accuracy of 85.0% (95% CI: 81.0%-90.0%), a specificity of 84.5% (95% CI: 78.9%-91.5%), a sensitivity of 86.7% (95% CI: 80.4%-93.3%) and an AUC of 0.928 (95% CI: 0.88-0.96). Deep CNNs can be an effective method to build low-budget embedded vision devices with limited computation power and memory capacity for diagnosis of oral cancer. Artificial intelligence (AI) can improve the quality and reach of oral cancer screening and early detection.

An intelligent work order classification model for government service based on multi-label neural network

The classification of work orders for livelihood service play an important role in the e-government system, which goal is to push these work orders to the corresponding processing department. Although many studies have realized the classification of work orders automatically, these work orders are either related to many subjects or related to many departments. Therefore, the problem of work order classification is not a single label classification problem, but a multi-label classification problem. In view of the continuous development of deep neural networks, the goal of this paper is to modify a MultilabelMarginLoss function and design a MultilabelCrossEntropyLoss function to realize the multi-classification system of work orders, which can push the work orders to the related departments(one or more departments) automatically according to their appeal contents (consisting of short texts). We perform deep neural network for text classification including TextCNN, TextRCNN, TextRNN, TextRNN_Att, FastText and DPCNN on the real government hotline data set and compare the performances of these models. The results reveal that TextCNN and TextRCNN have a more stable performance on different loss functions, especially on the MultilabelCrossEntropyLoss function.

Rapid and non-destructive spectroscopic method for classifying beef freshness using a deep spectral network fused with myoglobin information

A simple, novel, rapid, and non-destructive spectroscopic method that employs the deep spectral network for beef-freshness classification was developed. The deep-learning-based model classified beef freshness by learning myoglobin information and reflectance spectra over different freshness states. The reflectance spectra (480–920 nm) were measured from 78 beef samples for 17 days, and the datasets were sorted into three freshness classes based on their pH values. Myoglobin information showed statistically significant differences depending on the freshness; consequently, it was utilized as a crucial parameter for classification. The model exhibited improved performance when the reflectance spectra were combined with the myoglobin information. The accuracy of the proposed model improved to 91.9%, whereas that of the single-spectra model was 83.6%. Further, a high value for the area under the receiver operating characteristic curve (0.958) was recorded. This study provides a basis for future studies on the investigation of myoglobin information associated with meat freshness.

Identification of prognostic spatial organization features in colorectal cancer microenvironment using deep learning on histopathology images

BackgroundThe ecological diversity in the tumor microenvironment influences cancer progression and clinical outcomes of patients. However, the complexity of cellular and tissue components hamper quantitative dissection of the tumor microenvironment. In this study, we aimed to develop an efficient and robust artificial intelligence (AI)-empowered framework for the identification of prognostic spatial organization features based on histopathological images. ResultsUsing two public H&E image cohorts involving 107,180 hand-delineated image patches, we trained and validated a robust and efficient deep convolutional neural network for accurate tissue classification. With the classification result, we calculated whole-slide and infiltrating spatial organization features (SOFs) for different tissue types. Interestingly, the whole-slide SOFs recapitulated the characteristics of the four Consensus Molecular Subtypes (CMSs) of colorectal cancer (CRC). More specifically, we found that lymphocyte, tumor, mucus, and stroma tissues are significantly more abundant in CMS1, 2, 3, and 4, respectively. Using univariate and multivariate analyses, we identified infiltrating lymphocyte ratio (ILR) and infiltrating stroma ratio (ISR) are significantly associated with relapse-free survival. Based on two independent clinical cohorts, we further demonstrated the combinatorial prognostic value of ILR and ISR. Together, our results suggest that a high level of lymphocyte infiltration may surpass the effect of stromal infiltration. However, stromal infiltration will be essential for RFS in patients with low degrees of immune activity. ConclusionsWe developed CRC-SPA for accurate profiling of spatial organization features using histology images, providing a cost-efficient tool for more quantitative analysis of tumor microenvironment and stratification of patients for more optimized clinical management.

Wavelength Bands Reduction Method in Near-Infrared Hyperspectral Image based on Deep Neural Network for Tumor Lesion Classification

In this paper, we propose a method for wavelength bands reduction of near-infrared (NIR) hyperspectral imaging data to extract the cancer region with minimum input data. NIR hyperspectral imaging data has a spectrum data of each pixel and is suitable for distinguishing tumors region of the body rather than RGB imaging data. However, it is difficult to applicate to the medical field because of processing time consumption and hardware size limitation. Therefore, it is necessary to remove the redundant wavelength bands which are not (or little) contributed to tumor region extraction. Although several previous studies for wavelength bands reduction have been conducted, these approaches focused on the characteristics of the wavelength itself. In this research, the proposed wavelength bands reduction method is focused on the node weights of the post-training deep neural network which is an indicator directly related to classification. The experimental results using GIST specimen demonstrated that four wavelength bands selected from all wavelengths bands by using the proposed method are effective for tumor distinguishing as well as all wavelength bands.

Research on Surface Defect Detection of Solar Pv Panels Based on Pre-Training Network and Feature Fusion

Aiming at the problems of the lack of sample size and the complexity of defect images in defect detection task, based on the idea of transfer learning and hierarchical feature fusion, this paper proposes a deep classification network model of improved vgg19 pre-training network by analyzing the basic principle of feature extraction of convolutional neural network and getting inspiration from feature pyramid network. Then, the model is trained by the small-scale defect images of solar pv panel. Finally, the solar pv panel data set containing four kinds of defects, including cracks, debris, broken gates and black areas, is selected to comprehensively verify the effectiveness of the improved network in the defect detection task. The experimental results show that the proposed method is superior to the classical vgg19 network model in four evaluation indexes, such as accuracy, precision, recall rate and F1 score.

Deep convolutional neural network for preliminary in-field classification of lichen species

Lichens are unique organisms, valued for their pharmacological activity, but also well known as bioindicators of environmental pollution, key determinants for some natural ecological habitats, or just popular elements of decoration. High morphological similarity between lichen species makes their recognition complicated, especially under in-field conditions. Thus, there is a need for a quick and easy method that can help with the preliminary classification of selected lichen species. This paper presents a tool that can facilitate the recognition of Cladonia lichen species, based on a deep convolutional neural network, a model which has nowadays reached a classification level often comparable to humans. The network was trained and tested on twelve Cladonia species using a total of 1164 images, downloaded from various websites. The trained model achieved 60.94% accuracy, which is satisfactory for this novel, but still preliminary, automated classification of lichen species. • Deep convolutional neural network was designed to determine the lichen species. • The model was trained and tested on images of 12 Cladonia species, on1164 images. • Satisfactory 60.94% accuracy was achieved for the model. • This is the first use of neural network to identify lichen species. • The model can be useful for lichenologist and non-expert nature amateurs.

A Novel Multitemporal Deep Fusion Network (MDFN) for Short-Term Multitemporal HR Images Classification

High-resolution (HR) satellite images, due to the technical constraints on spectral and spatial resolutions, usually contain only several broad spectral bands but with a very high spatial resolution. This provides rich spatial details of the objects on the Earth surface, while their spectral discrimination is relatively low. Recently, the increase of the satellite revisit times made it possible to acquire more frequent data coverage for finer classification. In this article, we proposed a novel multitemporal deep fusion network (MDFN) for short-term multitemporal HR images classification. Specifically, a two-branch structure of MDFN is designed, which includes a long short-term memory (LSTM) and a convolutional neural network (CNN). The LSTM branch is mainly used to learn the joint expression of different temporal-spectral features. For the CNN branch, the three-dimensional (3-D) convolution is firstly applied along the temporal and spectral dimensions to jointly learn the temporal-spatial and spectral-spatial information, respectively, and then the 2-D convolution is performed along the spatial dimension to further extract the spatial context information. Finally, features generated from the two different branches are fused to obtain the discriminative high-level semantic information for classification. Experimental results carried on two real multitemporal HR remote sensing datasets demonstrate that the proposed MDFN provides better classification performance over the state-of-the-art methods, and it also shows the potentiality to use short-term multitemporal HR images for more accurate land use/land cover mapping.

Self-Supervised Deep Convolutional Neural Network for Chest X-Ray Classification

Chest radiography is a relatively cheap, widely available medical procedure that conveys key information for making diagnostic decisions. Chest X-rays are almost always used in the diagnosis of respiratory diseases such as pneumonia or the recent COVID-19. In this paper, we propose a self-supervised deep neural network that is pretrained on an unlabeled chest X-ray dataset. The learned representations are transferred to downstream task - the classification of respiratory diseases. The results obtained on four public datasets show that our approach yields competitive results without requiring large amounts of labeled training data.

A neoteric ensemble deep learning network for musculoskeletal disorder classification

A neoteric ensemble deep learning network for musculoskeletal disorder classification

Parallel implementation of content-guided deep convolutional network for hyperspectral image classification

Parallel implementation of content-guided deep convolutional network for hyperspectral image classification

Deep Learning Adapted to Differential Neural Networks Used as Pattern Classification of Electrophysiological Signals.

This manuscript presents the design of a deep differential neural network (DDNN) for pattern classification. First, we proposed a DDNN topology with three layers, whose learning laws are derived from a Lyapunov analysis, justifying local asymptotic convergence of the classification error and the weights of the DDNN. Then, an extension to include an arbitrary number of hidden layers in the DDNN is analyzed. The learning laws for this general form of the DDNN offer a contribution to the deep learning framework for signal classification with biological nature and dynamic structures. The DDNN is used to classify electroencephalographic signals from volunteers that perform an identification graphical test. The classification results show exponential growth in the signal classification accuracy from 82 percent with one layer to 100 percent with three hidden layers. Working with DDNN instead of static deep neural networks (SDNN) represents a set of advantages, such as processing time and training period reduction up to almost 100 times, and the increment of the classification accuracy while working with less hidden layers than working with SDNN, which are highly dependent on their topology and the number of neurons in each layer. The DDNN employed fewer neurons due to the induced feedback characteristic.

Classification of Remote Sensing Images Using EfficientNet-B3 CNN Model With Attention

Scene classification is a highly useful task in Remote Sensing (RS) applications. Many efforts have been made to improve the accuracy of RS scene classification. Scene classification is a challenging problem, especially for large datasets with tens of thousands of images with a large number of classes and taken under different circumstances. One problem that is observed in scene classification is the fact that for a given scene, only one part of it indicates which class it belongs to, whereas the other parts are either irrelevant or they actually tend to belong to another class. To address this issue, this paper proposes a deep attention Convolutional Neural Network (CNN) for scene classification in remote sensing. CNN models use successive convolutional layers to learn feature maps from larger and larger regions (or receptive fields) of the scene. The attention mechanism computes a new feature map as a weighted average of these original feature maps. In particular, we propose a solution, named EfficientNet-B3-Attn-2, based on the pre-trained EfficientNet-B3 CNN enhanced with an attention mechanism. A dedicated branch is added to layer 262 of the network, to compute the required weights. These weights are learned automatically by training the whole CNN model end-to-end using the backpropagation algorithm. In this way, the network learns to emphasize important regions of the scene and suppress the regions that are irrelevant to the classification. We tested the proposed EfficientNet-B3-Attn-2 on six popular remote sensing datasets, namely UC Merced, KSA, OPTIMAL-31, RSSCN7, WHU-RS19, and AID datasets, showing its strong capabilities in classifying RS scenes.