Proposed Convolutional Neural Network Research Articles

A convolution neural network for rapid and accurate staging of breast cancer based on mammography

IntroductionBreast cancer (BC) has been one of the main reasons for women's deaths in the recent decade. This study hypothesizes that the deep features resulting from mammography staging can be used to determine tumor staging and lymph nodes of the sentinel lymph node (SLN) before surgery in BC patients. MethodsA retrospective study of female BC patients being treated. Pathological variables were collected from medical records, including age, tumor location, staging based on the American Joint Committee on Cancer (AJCC), pathological type, human epidermal growth factor receptor 2 (HER2) status, Progesterone Receptor (PR) status, and Estrogen Receptor (ER) status. The contrast enhancement technique with Contrast-Limited Adaptive Histogram Equalization (CLAHE) was applied to the studied data using MATLAB software. The images were first changed from color to gray in the preprocessing process, and then the images were changed from gray to color again. The regions of interest (ROI) of the primary tumor were isolated from other areas of breast tissue by two expert radiologists. Therefore, a proposed convolution neural network (CNN) model was applied. ResultsIn this study, 390 patients' information was used to analyze the data. In total, out of 390 patients, 300 had metastatic SLN involvement, and 390 had a tumor size greater than 1 mm. Performance evaluation of the proposed CNN staging was AUC 0.919 (95 % CI: 0.852,0.971) for the training group and AUC 0.877 (95 % CI: 0.805,0.949) for the validation group. ConclusionsThis study proposes a deep learning model based on CNN based on mammographic images, which has a suitable function for determining tumor staging and SLN metastasis.

Adversarial Learning Improves Vision-Based Perception from Drones with Imbalanced Datasets

This work proposes a vision-based perception algorithm that combines image-processing-based detection and tracking of aerial objects with convolutional neural networks (CNNs) integrated for classification of general aviation aircraft, multirotor small uncrewed aerial systems (SUAS), fixed-wing SUAS, and birds to enable improved onboard avoidance algorithm decision making. Furthermore, we integrate adversarial learning during the training of the CNNs and evaluate performance with class balanced and imbalanced datasets because this maximizes the utility of resource-expensive flight experiments to collect aviation datasets. We compare our proposed CNN with adversarial learning (CNN+ADVL) model with a state-of-the-art CNN as well as a you only look once (YOLO, v4) model retrained (YOLO v4 aircraft) on the same data. The CNN+ADVL trained on the imbalanced dataset achieves the highest 10-fold cross-validation classification accuracy of 76.2% for aircraft and birds for all ranges while achieving 87.0% aircraft classification accuracy, meeting proposed self-assurance separation distances derived from Federal Aviation Administration (FAA) guidelines. In comparison, the CNNs achieved 74.4% 10-fold cross-validation classification accuracy for aircraft and birds as well as 83.4% accuracy for the aircraft, meeting proposed self-assurance separation distances derived from FAA guidelines. Furthermore, we demonstrate that the integration of adversarial learning improves the classification performance for the perception of aerial objects using a class imbalanced dataset.

A deep learning approach for multi‐stage classification of brain tumor through magnetic resonance images

AbstractBrain tumor is the 10th major cause of death among humans. The detection of brain tumor is a significant process in the medical field. Therefore, the objective of this research work is to propose a fully automated deep learning framework for multistage classification. Besides, this study focuses on to develop an efficient and reliable system using a convolutional neural network (CNN). In this study, the fast bounding box technique is used for segmentation. Moreover, the CNN layers‐based three models are developed for multistage classification through magnetic resonance images on three publicly available datasets. The first dataset is obtained from Kaggle Repository (Dataset‐1), the second dataset is known as Figshare (Dataset‐2), and the third dataset is called REMBRANDT (Dataset‐3) to classify the MR images into different grades. Different augmentation techniques are applied to increase the data size of MR images. In pre‐processing, the proposed models achieved higher Peak Signal‐to‐Noise ratio to remove noise. The first proposed deep CNN framework mentioned as Classification‐1 has obtained 99.40% accuracy, which classified MR images into two classes, that is (i) normal and (ii) abnormal, while the second proposed CNN framework mentioned as Classification‐2 has obtained 97.78% accuracy, which classified brain tumor into three types, which are meningioma, glioma, and pituitary. Similarly, the third developed CNN framework mentioned as Classification‐3 has obtained 98.91% accuracy that further classified MR images of tumors into four different classes as: Grade I, Grade II, Grade III, and Grade IV. The results demonstrate that the proposed models achieved better performance on three large and diverse datasets. The comparison of obtained outcomes shows that the developed models are more efficient and effective than state‐of‐the‐art methods.

Recognition and Classification of Handwritten Urdu Numerals Using Deep Learning Techniques

Urdu is a complex language as it is an amalgam of many South Asian and East Asian languages; hence, its character recognition is a huge and difficult task. It is a bidirectional language with its numerals written from left to right while script is written in opposite direction which induces complexities in the recognition process. This paper presents the recognition and classification of a novel Urdu numeral dataset using convolutional neural network (CNN) and its variants. We propose custom CNN model to extract features which are used by Softmax activation function and support vector machine (SVM) classifier. We compare it with GoogLeNet and the residual network (ResNet) in terms of performance. Our proposed CNN gives an accuracy of 98.41% with the Softmax classifier and 99.0% with the SVM classifier. For GoogLeNet, we achieve an accuracy of 95.61% and 96.4% on ResNet. Moreover, we develop datasets for handwritten Urdu numbers and numbers of Pakistani currency to incorporate real-life problems. Our models achieve best accuracies as compared to previous models in the literature for optical character recognition (OCR).

Computer-Aided Detection and Classification of Monkeypox and Chickenpox Lesion in Human Subjects Using Deep Learning Framework.

Monkeypox is a zoonotic viral disease caused by the monkeypox virus. After its recent outbreak, it has become clear that a rapid, accurate, and reliable diagnosis may help reduce the risk of a future outbreak. The presence of skin lesions is one of the most prominent symptoms of the disease. However, this symptom is also peculiar to chickenpox. The resemblance in skin lesions in the human subject may disrupt effective diagnosis and, as a result, lead to misdiagnosis. Such misdiagnosis can lead to the further spread of the disease as it is a communicable disease and can eventually result in an outbreak. As deep learning (DL) algorithms have recently been regarded as a promising technique in medical fields, we have been attempting to integrate a well-trained DL algorithm to assist in the early detection and classification of skin lesions in human subjects. This study used two open-sourced digital skin images for monkeypox and chickenpox. A two-dimensional convolutional neural network (CNN) consisting of four convolutional layers was applied. Afterward, three MaxPooling layers were used after the second, third, and fourth convolutional layers. Finally, we evaluated the performance of our proposed model with state-of-the-art deep-learning models for skin lesions detection. Our proposed CNN model outperformed all DL models with a test accuracy of 99.60%. In addition, a weighted average precision, recall, F1 score of 99.00% was recorded. Subsequently, Alex Net outperformed other pre-trained models with an accuracy of 98.00%. The VGGNet consisting of VGG16 and VGG19 performed least well with an accuracy of 80.00%. Due to the uniqueness of the proposed model and image augmentation techniques applied, the proposed CNN model is generalized and avoids over-fitting. This model would be helpful for the rapid and accurate detection of monkeypox using digital skin images of patients with suspected monkeypox.

Ethereum Cryptocurrency Entry Point and Trend Prediction using Bitcoin Correlation and Multiple Data Combination

Deep learning methods have achieved significant success in various applications, including trend signal prediction in financial markets. However, most existing approaches only utilize price action data. In this paper, we propose a novel system that incorporates multiple data sources and market correlations to predict the trend signal of Ethereum cryptocurrency. We conduct experiments to investigate the relationship between price action, candlestick patterns, and Ethereum-Bitcoin correlation, aiming to achieve highly accurate trend signal predictions. We evaluate and compare two different training strategies for Convolutional Neural Networks (CNNs), one based on transfer learning and the other on training from scratch. Our proposed 1-Dimensional CNN (1DCNN) model can also identify inflection points in price trends during specific periods through the analysis of statistical indicators. We demonstrate that our model produces more reliable predictions when utilizing multiple data representations. Our experiments show that by combining different types of data, it is possible to accurately identify both inflection points and trend signals with an accuracy of 98%.

COMPARATIVE ANALYSIS OF TRADITIONAL CLASSIFICATION AND DEEP LEARNING IN LUNG CANCER PREDICTION

The cancer is an intimidating illness. Extra care is necessary while making a diagnosis. To aid the identification process, medical imaging plays a crucial role by producing images of the internal organs of the body for better diagnosis of cancer. Medical images are typically utilized by radiologists, engineers, and clinicians to spot the inner constitution of either individual patients or group of individuals. Most doctors prefer computed tomography (CT) images for initial screening of cancer — mainly lung cancer. To achieve deeper understanding and categorization of lung cancer, diverse machine learning techniques are employed in image classification. Many research works have been done on the classification of CT images with different algorithms, but they failed to reach 100% accuracy. By applying methods like Support Vector Machine, deep learning system like artificial neural network (ANN) and proposed convolution neural network (CNN), a computerized system can be built for truthful classification. The models are built as a classification system that can identify the nodule, if present in the lungs, as benign, malignant or normal or as benign or normal. Lung cancer datasets at Iraq National Center aimed at Cancer Diseases (IQ-OTHNCCD) and Iran Hospital-based CT images are used in this research. SVM, ANN, and proposed CNN classification techniques are applied to the datasets considered. This research work, proposes a model for classification of CT images with very promising accuracy on the datasets considered.

Prediction of micro-hardness in thread rolling of St37 by convolutional neural networks and transfer learning.

The online version contains supplementary material available at 10.1007/s00170-022-10355-4.

Handwritten Geez Digit Recognition Using Deep Learning

Amharic language is the second most spoken language in the Semitic family after Arabic. In Ethiopia and neighboring countries more than 100 million people speak the Amharic language. There are many historical documents that are written using the Geez script. Digitizing historical handwritten documents and recognizing handwritten characters is essential to preserving valuable documents. Handwritten digit recognition is one of the tasks of digitizing handwritten documents from different sources. Currently, handwritten Geez digit recognition researches are very few, and there is no available organized dataset for the public researchers. Convolutional neural network (CNN) is preferable for pattern recognition like in handwritten document recognition by extracting a feature from different styles of writing. In this work, the proposed model is to recognize Geez digits using CNN. Deep neural networks, which have recently shown exceptional performance in numerous pattern recognition and machine learning applications, are used to recognize handwritten Geez digits, but this has not been attempted for Ethiopic scripts. Our dataset, which contains 51,952 images of handwritten Geez digits collected from 524 individuals, is used to train and evaluate the CNN model. The application of the CNN improves the performance of several machine-learning classification methods significantly. Our proposed CNN model has an accuracy of 96.21% and a loss of 0.2013. In comparison to earlier research works on Geez handwritten digit recognition, the study was able to attain higher recognition accuracy using the developed CNN model.

Deep learning-based approach for detecting COVID-19 in chest X-rays

Today, 2019 Coronavirus (COVID-19) infections are a major health concern worldwide. Therefore, detecting COVID-19 in X-ray images is crucial for diagnosis, evaluation, and treatment. Furthermore, expressing diagnostic uncertainty in a report is a challenging duty but unavoidable task for radiologists. This study proposes a novel CNN (Convolutional Neural Network) model for automatic COVID-19 identification utilizing chest X-ray images. The proposed CNN model is designed to be a reliable diagnostic tool for two-class categorization (COVID and Normal). In addition to the proposed model, different architectures, including the pre-trained MobileNetv2 and ResNet50 models, are evaluated for this COVID-19 dataset (13,824 X-ray images) and our suggested model is compared to these existing COVID-19 detection algorithms in terms of accuracy. Experimental results show that our proposed model identifies patients with COVID-19 disease with 96.71 percent accuracy, 91.89 percent F1-score. Our proposed approach CNN’s experimental results show that it outperforms the most advanced algorithms currently available. This model can assist clinicians in making informed judgments on how to diagnose COVID-19, as well as make test kits more accessible.

Winter wheat yield prediction using convolutional neural networks from environmental and phenological data

Crop yield forecasting depends on many interactive factors, including crop genotype, weather, soil, and management practices. This study analyzes the performance of machine learning and deep learning methods for winter wheat yield prediction using an extensive dataset of weather, soil, and crop phenology variables in 271 counties across Germany from 1999 to 2019. We proposed a Convolutional Neural Network (CNN) model, which uses a 1-dimensional convolution operation to capture the time dependencies of environmental variables. We used eight supervised machine learning models as baselines and evaluated their predictive performance using RMSE, MAE, and correlation coefficient metrics to benchmark the yield prediction results. Our findings suggested that nonlinear models such as the proposed CNN, Deep Neural Network (DNN), and XGBoost were more effective in understanding the relationship between the crop yield and input data compared to the linear models. Our proposed CNN model outperformed all other baseline models used for winter wheat yield prediction (7 to 14% lower RMSE, 3 to 15% lower MAE, and 4 to 50% higher correlation coefficient than the best performing baseline across test data). We aggregated soil moisture and meteorological features at the weekly resolution to address the seasonality of the data. We also moved beyond prediction and interpreted the outputs of our proposed CNN model using SHAP and force plots which provided key insights in explaining the yield prediction results (importance of variables by time). We found DUL, wind speed at week ten, and radiation amount at week seven as the most critical features in winter wheat yield prediction.

Efficient Malware Classification by Binary Sequences with One-Dimensional Convolutional Neural Networks

The rapid increase of malware attacks has become one of the main threats to computer security. Finding the best way to detect malware has become a critical task in cybersecurity. Previous work shows that machine learning approaches could be a solution to address this problem. Many proposed methods convert malware executables into grayscale images and apply convolutional neural networks (CNNs) for malware classification. However, converting malware executables into images could twist the one-dimensional structure of binary codes. To address this problem, we explore the bit and byte-level sequences from malware executables and propose efficient one-dimensional (1D) CNNs for the malware classification. Our experiments evaluate our proposed 1D CNN models with two benchmark datasets. Our proposed 1D CNN models achieve better performance from the experimental results than the existing 2D CNNs malware classification models by providing smaller resizing bit/byte-level sequences with less computational cost.

Power quality event classification using complex wavelets phasor models and customized convolution neural network

Origin and triggers of power quality (PQ) events must be identified in prior, in order to take preventive steps to enhance power quality. However it is important to identify, localize and classify the PQ events to determine the causes and origins of PQ disturbances. In this paper a novel algorithm is presented to classify voltage variations into six different PQ events considering the space phasor model (SPM) diagrams, dual tree complex wavelet transforms (DTCWT) sub bands and the convolution neural network (CNN) model. The input voltage data is converted into SPM data, the SPM data is transformed using 2D DTCWT into low pass and high pass sub bands which are simultaneously processed by the 2D CNN model to perform classification of PQ events. In the proposed method CNN model based on Google Net is trained to perform classification of PQ events with default configuration as in deep neural network designer in MATLAB environment. The proposed algorithm achieve higher accuracy with reduced training time in classification of events than compared with reported PQ event classification methods.

Deep learning-based automatic detection of tuberculosis disease in chest X-ray images.

PurposeTo train a convolutional neural network (CNN) model from scratch to automatically detect tuberculosis (TB) from chest X-ray (CXR) images and compare its performance with transfer learning based technique of different pre-trained CNNs.Material and methodsWe used two publicly available datasets of postero-anterior chest radiographs, which are from Montgomery County, Maryland, and Shenzhen, China. A CNN (ConvNet) from scratch was trained to automatically detect TB on chest radiographs. Also, a CNN-based transfer learning approach using five different pre-trained models, including Inception_v3, Xception, ResNet50, VGG19, and VGG16 was utilized for classifying TB and normal cases from CXR images. The performance of models for testing datasets was evaluated using five performances metrics, including accuracy, sensitivity/recall, precision, area under curve (AUC), and F1-score.ResultsAll proposed models provided an acceptable accuracy for two-class classification. Our proposed CNN architecture (i.e., ConvNet) achieved 88.0% precision, 87.0% sensitivity, 87.0% F1-score, 87.0% accuracy, and AUC of 87.0%, which was slightly less than the pre-trained models. Among all models, Exception, ResNet50, and VGG16 provided the highest classification performance of automated TB classification with precision, sensitivity, F1-score, and AUC of 91.0%, and 90.0% accuracy.ConclusionsOur study presents a transfer learning approach with deep CNNs to automatically classify TB and normal cases from the chest radiographs. The classification accuracy, precision, sensitivity, and F1-score for the detection of TB were found to be more than 87.0% for all models used in the study. Exception, ResNet50, and VGG16 models outperformed other deep CNN models for the datasets with image augmentation methods.

AI based colorectal disease detection using real-time screening colonoscopy.

Colonoscopy is an effective tool for early screening of colorectal diseases. However, the application of colonoscopy in distinguishing different intestinal diseases still faces great challenges of efficiency and accuracy. Here we constructed and evaluated a deep convolution neural network (CNN) model based on 117 055 images from 16 004 individuals, which achieved a high accuracy of 0.933 in the validation dataset in identifying patients with polyp, colitis, colorectal cancer (CRC) from normal. The proposed approach was further validated on multi-center real-time colonoscopy videos and images, which achieved accurate diagnostic performance on detecting colorectal diseases with high accuracy and precision to generalize across external validation datasets. The diagnostic performance of the model was further compared to the skilled endoscopists and the novices. In addition, our model has potential in diagnosis of adenomatous polyp and hyperplastic polyp with an area under the receiver operating characteristic curve of 0.975. Our proposed CNN models have potential in assisting clinicians in making clinical decisions with efficiency during application.

A Hybrid Deep CNN Model for Abnormal Arrhythmia Detection Based on Cardiac ECG Signal.

Electrocardiogram (ECG) signals play a vital role in diagnosing and monitoring patients suffering from various cardiovascular diseases (CVDs). This research aims to develop a robust algorithm that can accurately classify the electrocardiogram signal even in the presence of environmental noise. A one-dimensional convolutional neural network (CNN) with two convolutional layers, two down-sampling layers, and a fully connected layer is proposed in this work. The same 1D data was transformed into two-dimensional (2D) images to improve the model’s classification accuracy. Then, we applied the 2D CNN model consisting of input and output layers, three 2D-convolutional layers, three down-sampling layers, and a fully connected layer. The classification accuracy of 97.38% and 99.02% is achieved with the proposed 1D and 2D model when tested on the publicly available Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) arrhythmia database. Both proposed 1D and 2D CNN models outperformed the corresponding state-of-the-art classification algorithms for the same data, which validates the proposed models’ effectiveness.

4D deep image prior: dynamic PET image denoising using an unsupervised four-dimensional branch convolutional neural network

Although convolutional neural networks (CNNs) demonstrate the superior performance in denoising positron emission tomography (PET) images, a supervised training of the CNN requires a pair of large, high-quality PET image datasets. As an unsupervised learning method, a deep image prior (DIP) has recently been proposed; it can perform denoising with only the target image. In this study, we propose an innovative procedure for the DIP approach with a four-dimensional (4D) branch CNN architecture in end-to-end training to denoise dynamic PET images. Our proposed 4D CNN architecture can be applied to end-to-end dynamic PET image denoising by introducing a feature extractor and a reconstruction branch for each time frame of the dynamic PET image. In the proposed DIP method, it is not necessary to prepare high-quality and large patient-related PET images. Instead, a subject’s own static PET image is used as additional information, dynamic PET images are treated as training labels, and denoised dynamic PET images are obtained from the CNN outputs. Both simulation with [18F]fluoro-2-deoxy-D-glucose (FDG) and preclinical data with [18F]FDG and [11C]raclopride were used to evaluate the proposed framework. The results showed that our 4D DIP framework quantitatively and qualitatively outperformed 3D DIP and other unsupervised denoising methods. The proposed 4D DIP framework thus provides a promising procedure for dynamic PET image denoising.

Exploring Convolutional Neural Networks Architectures for the Classification of Hand-Drawn Shapes in Learning Therapy Applications

A positive consequence of the existence of a more inclusive society is the appearance of protocols to help identify those in need. One of these protocols is the application of tests with the aim to detect learning disabilities in children so that opportune intervention could be made. These advances, though, also pose challenges to the responsible to administer these tests, for example, the large number of tests to evaluate that make the process lengthy at the best. In this study, we implement a Convolutional Neural Network (CNN) model for the automatic classification of hand-drawn drawings of the Bender Gestalt Test (BGT), which is a test that evaluates the perceptual-motor maturity and perceptual disorder on individuals. In BGT, nine different drawings are presented to the patient who must reproduce them using pencil and paper. This study focuses on the automatic detection of the traces and their classification, then aiming to expedite the test evaluation process. Our proposed task-specialized CNN, named CNN4-Bender, is compared against other eleven neural-network based models registering an average performance of 91.56%, surpassed only by ResNet50 but with a high computational cost in this last one. To further evaluate our model we also consider other classification tasks that include the MNIST and OIHACDB datasets, where the CNN4-Bender architecture obtains a competitive performance and in some cases outperforms state-of-the-art models.

Potato Leaf Disease Diagnosis and Detection System Based on Convolution Neural Network

For decades, agriculture has been an essential food source. According to related statics, over 60% of the total earth population mainly depend on agriculture’s sources for their primary feed. Unfortunately, one of the disaster problems that affect badly on agriculture production is plant diseases. There are about 25% of agriculture production lost annually because of plant diseases. Late and Early Blight diseases are one of the most destructive diseases that infect potato crop. Although, the late and inaccurate detection of plant diseases increases the losing percentage for the crop. The main approach of our proposed system is to detect early the plant diseases to decrease the plant’s production losses by using a diagnosis and detection system based on the Convolution Neural Network (CNN). We used CNN to extract the diseases features from the input images of the supported training dataset for classification purposes. For model training, 1700 of potato leaf images were used, then the testing process is done by using approximately 300 images and 100 images for fine tuning and parameters calibration against any biased data. Our proposed CNN architecture archives 98.2% accuracy, which is higher compared with other approaches run on the same dataset.

Bangla-Meitei Mayek scripts handwritten character recognition using Convolutional Neural Network

Recognition of handwritten characters in two Indic scripts Bangla and Meitei Mayek is one of the challenging responsibilities due to intricate patterns and scarcity of standard datasets. Convolutional Neural Network (CNN) is one of the stablest well-known techniques for classifying objects in distinctive specialties as it has an extraordinary capability of discovering complex patterns. In this paper, we hook a different layout and obtain a unique CNN architecture from scratch, which has manifold advantages over classical machine learning (ML) approaches, and it has a unique ability to consolidate feature extraction and classification altogether. Further, we stretch our work to uncover the mathematical rationale for using non-linearity in the deep learning (DL) model. Our proposed CNN architecture consists of four layers, including convolutional layer (CL), nonlinear activation layer (AL), pooling layer (PL), and fully connected layer (FCL), which are used in the existing two accessible Bangla datasets named cMATERdb and ISI Bangla datasets. The identical model also validates on proposed Manipuri Character dataset, called “Mayek27”. Moreover, we perform an in-depth comparison with different batch sizes and optimization techniques over all the datasets for understanding their functionality. We conceive a novel benchmark performance that has delivered state-of-the-art decisions on two regional handwritten character identifications.