High Classification Results Research Articles

Comparison of Explainable Artificial Intelligence Model and Radiologist Review Performances to Detect Breast Cancer in 752 Patients.

Breast cancer is a type of cancer caused by the uncontrolled growth of cells in the breast tissue. In a few cases, erroneous diagnosis of breast cancer by specialists and unnecessary biopsies can lead to various negative consequences. In some cases, radiologic examinations or clinical findings may raise the suspicion of breast cancer, but subsequent detailed evaluations may not confirm cancer. In addition to causing unnecessary anxiety and stress to patients, such diagnosis can also lead to unnecessary biopsy procedures, which are painful, expensive, and prone to misdiagnosis. Therefore, there is a need for the development of more accurate and reliable methods for breast cancer diagnosis. In this study, we proposed an artificial intelligence (AI)-based method for automatically classifying breast solid mass lesions as benign vs malignant. In this study, a new breast cancer dataset (Breast-XD) was created with 791 solid mass lesions belonging to 752 different patients aged 18 to 85 years, which were examined by experienced radiologists between 2017 and 2022. Six classifiers, support vector machine (SVM), K-nearest neighbor (K-NN), random forest (RF), decision tree (DT), logistic regression (LR), and XGBoost, were trained on the training samples of the Breast-XD dataset. Then, each classifier made predictions on 159 test data that it had not seen before. The highest classification result was obtained using the explainable XGBoost model (X2GAI) with an accuracy of 94.34%. An explainable structure is also implemented to build the reliability of the developed model. The results obtained by radiologists and the X2GAI model were compared according to the diagnosis obtained from the biopsy. It was observed that our developed model performed well in cases where experienced radiologists gave false positive results.

Polar Sea Ice Monitoring Using HY-2B Satellite Scatterometer and Scanning Microwave Radiometer Measurements

The Ku band microwave scatterometer (SCA) and scanning microwave radiometer (SMR) onboard HaiYang-2B (HY-2B) can simultaneously supply active and passive microwave observations over the polar region. In this paper, a polar ice water discrimination model and Arctic sea-ice-type classification model based on the support vector machine (SVM) method were established and used to produce a daily sea ice extent dataset from 2019 to 2021 with data from SCA and SMR. First, suitable scattering and radiation parameters are chosen as input data for the discriminant model. Then, the sea ice extent was obtained based on the monthly ice water discrimination model, and finally, the ice over the Arctic was classified into multiyear ice (MYI) and first-year ice (FYI). The 3-year ice extent and MYI extent products were consistent with the similar results of the National Snow and Ice Data Center (NSIDC) and Ocean and Sea Ice Satellite Application Facility (OSISAF). Using the OSISAF similar product as validation data, the overall accuracies (OAs) of ice/water discrimination and FYI/MYI discrimination are 99% and 97%, respectively. Compared with the high spatial resolution classification results of the Moderate Resolution Imaging Spectroradiometer (MODIS) and SAR, the OAs of ice/water discrimination and FYI/MYI discrimination are 96% and 86%, respectively. In conclusion, the SAC and SMR of HY-2B have been verified for monitoring polar sea ice, and the sea ice extent and sea-ice-type products are promising for integration into long-term sea ice records.

Towards ultrasonic guided wave fine-grained damage detection on hierarchical multi-label classification network

This study advocates the hierarchical multi-label classification (HMC) network for fine-grained damage detection using ultrasonic guided wave. Existing deep learning methods only focus on one aspect of the damage type or the degree of damage progression, and both are important for the long-term safe operation of the structure. To address this limitation, an ultrasonic guided wave HMC network (GHmcNet) and its lightweight version (L-GHmcNet) are developed. The essential designs of the proposed methods are derived from two motivations: (1) achieving robust fine-grained damage classification; (2) designing a lightweight network. The hierarchical labels (defect type-depth-size) are constructed on guided wave data. Inspired by the work of image classification, residual connections are introduced to transfer and fuse features between parent and children categories. The GHmcNet has three output channels and a weighted cross-entropy loss function is imposed on each channel. The weights of the corresponding loss are automatically adjusted during training. To reduce the parameters, we improve the structure of GHmcNet by applying different numbers of convolutional layers and channels at different levels based on the number of classes in each level, proposing the lightweight model L-GHmcNet. Both numerical and experimental studies are carried out, in which the dataset contains signals of different damage types, depths and sizes. The results show the superiority of the proposed methods in terms of accuracy at three levels compared to state-of-the-art methods. Moreover, the high accuracy classification results of L-GHmcNet demonstrate that the proposed lightweight network is a successful attempt.

Automatic classification of emotions in speech: methods and data

The subject of this study is the data and methods used in the task of automatic recognition of emotions in spoken speech. This task has gained great popularity recently, primarily due to the emergence of large datasets of labeled data and the development of machine learning models. The classification of speech utterances is usually based on 6 archetypal emotions: anger, fear, surprise, joy, disgust and sadness. Most modern classification methods are based on machine learning and transformer models using a self-learning approach, in particular, models such as Wav2vec 2.0, HuBERT and WavLM, which are considered in this paper. English and Russian datasets of emotional speech, in particular, the datasets Dusha and RESD, are analyzed as data. As a method, an experiment was conducted in the form of comparing the results of Wav2vec 2.0, HuBERT and WavLM models applied to the relatively recently collected Russian datasets of emotional speech Dusha and RESD. The main purpose of the work is to analyze the availability and applicability of available data and approaches to recognizing emotions in speech for the Russian language, for which relatively little research has been conducted up to this point. The best result was demonstrated by the WavLM model on the Dusha dataset - 0.8782 dataset according to the Accuracy metric. The WavLM model also received the best result on the RESD dataset, while preliminary training was conducted for it on the Dusha - 0.81 dataset using the Accuracy metric. High classification results, primarily due to the quality and size of the collected Dusha dataset, indicate the prospects for further development of this area for the Russian language.

Intrusion Detection Using Bat Optimization Algorithm and DenseNet for IoT and Cloud Based Systems

In Internet of Things (IoT) and cloud systems, Intrusion Detection (ID) is very vital for protecting the security infrastructures. ID techniques are extensively used to detect and track malicious threats in cloud and IoT systems. In the IoT based ID, the conventional techniques work based on the manual traffic feature values that increase the complexity of the networks and achieve a limited detection rate on the larger IoT databases. For addressing the above-stated issues and achieving high classification results, an effective deep learning based ID-System (IDS) is implemented in this article. Initially, the IoT data is acquired from the NSW-NB15 and NSL-KDD databases, and then, the standard scaling normalization technique, known as Min-Max normalization, is applied to select the dominant attributes and to eliminate outliers from the acquired databases. Additionally, the optimal features are selected from the rescaled normalized data by implementing the Bat optimization algorithm. The selection of optimal features decreases the computational complexity and training time of the IDS. The chosen optimal features are passed into the DenseNet model for carrying out intrusion attack detection. Particularly, in the binary-class classification, the Bat-based DenseNet model obtained 98.89% and 98.40% of accuracy on the UNSW-NB15 and NSL-KDD databases, correspondingly. The obtained simulation results prove the higher effectiveness of the current study when it is related to the state-of-the-art classifiers.

CLASSIFICATION OF URBAN LAND USE AND LAND COVER WITH K-NEAREST NEIGHBOUR CLASSIFIER IN THE CITY OF CAPE TOWN, SOUTH AFRICA – CAPE FLATS CASE STUDY

Abstract. The rapid growth of cities, owing to rural-urban migration and high birth rate has resulted in encroachment of the land use and land cover (LULC) in the Cape Flats, situated north of Cape Town, Western Cape, South Africa. PlanetScope imagery and K-Nearest Neighbour (KNN) classifier are used to map and monitor the following LULC classes within the Cape Flats; waterbodies, trees, built-up, and vegetation, we further do a time series map between 2016 and 2021.The results showed that LULC changes between 2016 to 2021 are as follows: a decrease of 2.1% in vegetation, 0.4% in water bodies, and 11% in trees. Built-up areas, on the other hand, showed a significant increase of 13.6% over five years. The LULC changes in the Cape Flats were mainly triggered by an increase in built-up areas due to household construction to accommodate the increased population resulting from rural-to-urban migration and high birth rate. Classification accuracy from 2016 and 2021 was as follows; overall accuracy of 98.31% and 0.97 kappa coefficient in 2016, while overall accuracy 96.54% and kappa coefficient 0.95 in 2021. A combination of machine learning and high-resolution imagery showcased that high classification results can be achieved in monitoring subtle LULC changes. We recommended that all relevant stakeholders, including government officials and municipalities, formulate and adopt policies to protect against LULC degradation.

The impact of (simulated) resolution on breast cancer diagnosis based on high-resolution 3D micro-CT microcalcification images.

Breast microcalcifications (MCs) are considered to be a robust marker of breast cancer. A machine learning model can provide breast cancer diagnosis based on properties of individual MCs - if their characteristics are captured at high resolution and in3D. The main purpose of the study was to explore the impact of image resolution (8µm, 16µm, 32µm, 64µm) when diagnosing breast cancer using radiomics features extracted from individual high resolution 3D micro-CT MCimages. Breast MCs extracted from 86 female patients were analyzed at four different spatial resolutions: 8µm (original resolution) and 16µm, 32µm, 64µm (simulated image resolutions). Radiomic features were extracted at each image resolution in an attempt, to find a compact feature signature allowing to distinguish benign and malignant MCs. Machine learning algorithms were used for classifying individual MCs and samples (i.e., patients). For sample diagnosis, a custom-based thresholding approach was used to combine individual MC results into sample results. We conducted classification experiments when using (a) the same MCs visible in 8µm, 16µm, 32µm, and 64µm resolution; (b) the same MCs visible in 8µm, 16µm, and 32µm resolution; (c) the same MCs visible in 8µm and 16µm resolution; (d) all MCs visible in 8µm, 16µm, 32µm, and 64µm resolution. Accuracy, sensitivity, specificity, AUC, and F1 score were computed for eachexperiment. The individual MC results yielded an accuracy of 77.27%, AUC of 83.83%, F1 score of 77.25%, sensitivity of 80.86%, and specificity of 72.2% at 8µm resolution. For the individual MC classifications we report for the F1 scores: a 2.29% drop when using 16µm instead of 8µm, a 4.01% drop when using 32µm instead of 8µm, a 10.69% drop when using 64µm instead of 8µm. The sample results yielded an accuracy and F1 score of 81.4%, sensitivity of 80.43%, and specificity value of 82.5% at 8µm. For the sample classifications we report for F1 score values: a 6.3% drop when using 16µm instead of 8µm, a 4.91% drop when using 32µm instead of 8µm, and a 6.3% drop when using 64µm instead of 8µm. The highest classification results are obtained at the highest resolution (8µm). If breast MCs characteristics could be visualized/captured in 3D at a higher resolution compared to what is used nowadays in digital mammograms (approximately 70µm), breast cancer diagnosis will beimproved.

Optimizing Deep Learning Model Parameters Using Socially Implemented IoMT Systems for Diabetic Retinopathy Classification Problem

Diabetic retinopathy (DR) is on the increase nowadays due to the high sugar level in the blood, and it is the reason for blindness that mainly occurs in middle-aged people. Furthermore, the Internet of Medical Things (IoMT) enabled computer-aided diagnostic (CAD) systems record DR-related data online and give patients with reassuring information. The internet allows for the interconnection of a variety of smart devices, enabling remote healthcare systems based on the IoMT to link patients with medical professionals. The proper diagnosis of diabetic patients and detection of DR severity in earlier stages help in preventing blindness. Therefore, the basic aim of this study is to prevent the diabetic patient from losing vision by detecting and classifying the severity of DR fundus images using the IoMT-enabled CAD system. This article designed a novel diabetic retinopathy classification (DRC) system by hybridizing the DL model with optimization algorithms to classify the DR images based on severity. This system begins with preprocessing phase for removing the noise from edges. Next, the proposed K-mean cluster-based growing region segmentation is employed to extract the useful region from the images. Then, pretrained convolutional neural network (CNN) model, i.e., RESnet with the proposed hybrid genetic and ant colony optimization (HGACO) algorithm, is applied to extract the features from the region of interest (ROI) and classify them into four severity levels. Performance indices such as AUC, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$F$</tex-math> </inline-formula> -measure, accuracy, sensitivity, and specificity are analyzed to evaluate the performance on MESSIDOR dataset. The performance of the proposed DRC system is compared with state-of-the-art classification systems. The proposed HGACO algorithm is also compared with Adam and gradient descent (GD) optimizers. The system is also evaluated by employing different parameters of CNN and HGACO. Additionally, for the determination of the classification accuracy of the DRC system, the confidence interval statistical test is implemented considering various parameters and configurations of the neural network. The results revealed that the proposed DR system provides higher classification results by achieving 95.78%, 91.98%, 97.7%, and 94.56% AUC, sensitivity, accuracy, and specificity rate, respectively. CNN alleviates the difficulty of developing image features, whereas the HGACO algorithm-based technique automates CNN hyperparameter design.

Question answering models for human–machine interaction in the manufacturing industry

This paper presents a question answering (QA) system that will enable workers from the manufacturing industry to ’hands-free’ request information. This kind of systems, that are broadly used in household context, have started to gain popularity in industrial environments. To develop the system, PDF-based industrial manuals in Spanish have been processed, annotated by experts and used to train the different components of the architecture, i.e. a question classifier and two QA systems. Different metrics have been applied and developed in order to test the performance of the system. The developed architecture obtains high classification and correct response results, demonstrating the viability of these systems in the industry.

IMPROVING CLASSIFICATION PERFORMANCE USING ENSEMBLE LEARNING APPROACH

The data mining techniques are used for evaluation of the data in order to find and represent the data in such manner by which the applications are becomes beneficial. Therefore, different kinds of computational algorithms and modeling’s are incorporated for analyzing the data. These computational algorithms are help to understand the data patterns and their application utility. The data mining algorithms supports supervised as well as unsupervised techniques of data analysis. This work is aimed to investigate about the supervised learning technique specifically performance improvements on classification techniques. The proposed classification model includes the multiple classifiers namely Bayesian classifier, k-nearest neighbor and the c4.5 decision tree algorithm. By nature of the outcomes and the modeling of the data these algorithms are functioning differently from each other. Thus, a weight based classification technique is introduced in this work. The weight is a combination of outcomes provided by the implemented three classifiers in terms of their predicted class labels. Using the weighted outcomes, the final class label for the input data instance is decided. The implementation of the proposed working model is performed with the help of JAVA and WEKA classes. The results obtained by experimentation of the proposed approach with the vehicle data set demonstrate the high accurate classification results. Thus, the proposed model is an effective classification technique as compared to single model implementation for classification task.

Fast and effective classification of plastic waste by pushbroom hyperspectral sensor coupled with hierarchical modelling and variable selection

Plastic waste management represents a global challenge in the framework of sustainable production and consumption of resources. One of the most critical issues in plastic recycling is polymer separation, necessary to obtain high-quality secondary raw material flow streams. The aim of this work was to build a classification strategy, based on pushbroom hyperspectral imaging, able to recognize the most common polymers found in mixed plastic waste to be applied at recycling plant scale. After exploring polymer spectral differences by principal component analysis, a hierarchical partial least squares-discriminant analysis, based on the acquired full spectra, and a hierarchical interval partial least squares-discriminant analysis, based on selected variables, were tested and their performances were evaluated and compared. High quality classification results were obtained in both cases, demonstrating that the developed multi-class models can be utilized in a flexible way for quality control and/or for on-line sorting actions in recycling plants.

A comparative study of several classification metrics and their performances on data

Six classification metrics namely, Accuracy, Precision, Recall (Sensitivity), Specificity, F1-Score and Area Under the Curve have been studied in this work. A classification model based on the Support Vector Machine, was used to obtain a confusion matrix, which provided the needed information for calculating the different classification metrics. Twenty different datasets were used to assess the performances of the classification metrics. Accuracy and Area Under the Curve are the two metrics that consistently gave a classification result given each dataset used in the study. Although accuracy appears to be marginally better that AUC, it was discovered that in some cases where sensitivity is zero, accuracy yielded a high correct classification result. This goes further to implying that prior to choosing accuracy as a preferred metric for classification, investigation should be carried out to find out what sensitivity and specificity are. Where there are high values for sensitivity and specificity, the study shows that a choice of accuracy as a preferred classification metric leads to a high percentage of correct classification result.

Lightweight Classification Network for Pulmonary Tuberculosis Based on CT Images

With the continuous development of medical informatics and digital diagnosis, the classification of tuberculosis cases from computed tomography (CT) images of the lung based on deep learning is an important guiding aid in clinical diagnosis and treatment. Due to its potential application in medical image classification, this task has received extensive research attention. Existing related neural network techniques are still challenging in terms of feature extraction of global contextual information of images and network complexity in achieving image classification. To address these issues, this paper proposes a lightweight medical image classification network based on a combination of Transformer and convolutional neural network (CNN) for the classification of tuberculosis cases from lung CT. The method mainly consists of a fusion of the CNN module and the Transformer module, exploiting the advantages of both in order to accomplish a more accurate classification task. On the one hand, the CNN branch supplements the Transformer branch with basic local feature information in the low level; on the other hand, in the middle and high levels of the model, the CNN branch can also provide the Transformer architecture with different On the other hand, in the middle and high levels of the model, the CNN branches can also provide different local and global feature information to the Transformer architecture to enhance the ability of the model to obtain feature information and improve the accuracy of image classification. A shortcut is used in each module of the network to solve the problem of poor model results due to gradient divergence and to optimise the effectiveness of TB classification. The proposed lightweight model can well solve the problem of long training time in the process of TB classification of lung CT and improve the speed of classification. The proposed method was validated on a computed tomography (CT) image dataset provided by the First Hospital of Lanzhou University. The experimental results show that the proposed lightweight classification network for tuberculosis based on CT medical images of lungs can fully extract the feature information of the input images and obtain high accuracy classification results.

Analysis of electrophysiological and mechanical dimensions of swallowing by non-invasive biosignals

Objective:Alterations in the neuromuscular coordination of swallowing are known as dysphagia, which can produce malnutrition, dehydration and aspiration pneumonia. Its instrumental diagnosis is invasive and expertise dependent. Thus, we introduced a non-invasive multimodal approach for dysphagia screening using surface electromyography (sEMG) and accelerometry-based cervical auscultation (Acc). Methods:Thirty healthy individuals and 30 patients with functional oropharyngeal dysphagia were recruited. Swallowing tasks of saliva and 5, 10, and 20mL of yogurt and water were performed. Supra- and infrahyoid sEMG and tri-axial Acc signals were recorded. Linear and non-linear features were extracted and selected. Two unimodal and one multimodal classification scenarios were tested. Classical algorithms were applied and the Area Under the ROC curve (AUC) was the criterion for hyperparameters optimization. Results:The Acc related features were the most consistently selected. Although the classification results with Acc signals were higher than with sEMG, the signal fusion improved the unimodal results regardless of swallowing task (AUC > 0.82). The highest classification results were achieved with small volumes of water (AUC = 0.86 ± 0.15) and yogurt (AUC = 0.87 ± 0.12). Conclusion:The combination of non-invasive sEMG and Acc signals improves the performance of automatic classification models for dysphagia detection. Significance:This paper proposes a multimodal approach based on electrophysiological and mechanical swallowing dimensions, for automatic, non-invasive and quantitative dysphagia screening.

Identifying peanut maturity based on Hyper Spectral Invariant Scaled Feature Selection using Adaptive Dense Net Recurrent Neural Network

With the rapid development of agricultural resources in south India, the peanut is important for cultivation in delta regions. The findings of suitable peanut seeds are analyzed through digital image processing with the support of deep learning analysis. Identifying seed maturity is a challenging task because seed level finding in features contains various scaled our dimensions, and increasing on successive feature observation leads to maturity findings inaccuracy in precision rate. We propose a Hyper Spectral Invariant Scaled Feature Selection (HSISFS) using an Adaptive Dense Net Recurrent Neural Network (ADNRNN) to resolve this problem. This initiates with preprocessing and scaling peanut entities' regions to normalize the image pixel by reducing the noiseless image. The entries of the peanut image features are observed by covering the exact boundary regions to apply the Space Retention Log Edge Detector (SRLogED) to make sliding segmentation. The sliding segments are normalized to find the matured objects through color quantization by observing pods entities. This is enhanced with Entity scalar histogram equalization (ESHE) by projecting the entities of peanut buds regions to find the scaled weights of features. Then the equivalent distance of matured and non-matured seeds are extracted through the hyperspectral Invariant scaled feature selection method and classified using the Adaptive Dense Net Recurrent Neural Network Algorithm. This proposed system produces high classification results by finding the exact scaled feature dimension to predict the maturity of seed and higher performance to increase the precision rate compared to the other systems.

Match them up: visually explainable few-shot image classification

Few-shot learning (FSL) approaches, mostly neural network-based, assume that pre-trained knowledge can be obtained from base (seen) classes and transferred to novel (unseen) classes. However, the black-box nature of neural networks makes it difficult to understand what is actually transferred, which may hamper FSL application in some risk-sensitive areas. In this paper, we reveal a new way to perform FSL for image classification, using a visual representation from the backbone model and patterns generated by a self-attention based explainable module. The representation weighted by patterns only includes a minimum number of distinguishable features and the visualized patterns can serve as an informative hint on the transferred knowledge. On three mainstream datasets, experimental results prove that the proposed method can enable satisfying explainability and achieve high classification results. Code is available at https://github.com/wbw520/MTUNet.

Automated Diabetic Retinopathy Detection Using Horizontal and Vertical Patch Division-Based Pre-Trained DenseNET with Digital Fundus Images.

Diabetic retinopathy (DR) is a common complication of diabetes that can lead to progressive vision loss. Regular surveillance with fundal photography, early diagnosis, and prompt intervention are paramount to reducing the incidence of DR-induced vision loss. However, manual interpretation of fundal photographs is subject to human error. In this study, a new method based on horizontal and vertical patch division was proposed for the automated classification of DR images on fundal photographs. The novel sides of this study are given as follows. We proposed a new non-fixed-size patch division model to obtain high classification results and collected a new fundus image dataset. Moreover, two datasets are used to test the model: a newly collected three-class (normal, non-proliferative DR, and proliferative DR) dataset comprising 2355 DR images and the established open-access five-class Asia Pacific Tele-Ophthalmology Society (APTOS) 2019 dataset comprising 3662 images. Two analysis scenarios, Case 1 and Case 2, with three (normal, non-proliferative DR, and proliferative DR) and five classes (normal, mild DR, moderate DR, severe DR, and proliferative DR), respectively, were derived from the APTOS 2019 dataset. These datasets and these cases have been used to demonstrate the general classification performance of our proposal. By applying transfer learning, the last fully connected and global average pooling layers of the DenseNet201 architecture were used to extract deep features from input DR images and each of the eight subdivided horizontal and vertical patches. The most discriminative features are then selected using neighborhood component analysis. These were fed as input to a standard shallow cubic support vector machine for classification. Our new DR dataset obtained 94.06% and 91.55% accuracy values for three-class classification with 80:20 hold-out validation and 10-fold cross-validation, respectively. As can be seen from steps of the proposed model, a new patch-based deep-feature engineering model has been proposed. The proposed deep-feature engineering model is a cognitive model, since it uses efficient methods in each phase. Similar excellent results were seen for three-class classification with the Case 1 dataset. In addition, the model attained 87.43% and 84.90% five-class classification accuracy rates using 80:20 hold-out validation and 10-fold cross-validation, respectively, on the Case 2 dataset, which outperformed prior DR classification studies based on the five-class APTOS 2019 dataset. Our model attained about >2% classification results compared to others. These findings demonstrate the accuracy and robustness of the proposed model for classification of DR images.

A Novel Deep Learning-Based Bidirectional Elman Neural Network for Facial Emotion Recognition

Facial emotion recognition (FER) is an interesting area of research. It has a wide range of applications, but there is still a deficiency of an accurate approach to provide better results. A novel FER system to maximize classification accuracy has been introduced in this paper. The proposed approach constitutes the following phases: pre-processing, feature extraction, feature selection, and classification. Initially, the images are pre-processed using the extended cascaded filter (ECF) and then the geometric and appearance-based features are extracted. An enhanced battle royale optimization (EBRO) for feature selection has been proposed to select the relevant features and to reduce the dimensionality problem. Then, the classification is carried out using a novel bidirectional Elman neural network (Bi-ENN) that offers high classification results. The proposed Bi-ENN-based emotion classification can accurately discriminate the input features. It enabled the model to predict the labels for classification accurately. The proposed model on evaluations attained an accuracy rate of 98.57% on JAFFE and 98.75% on CK+ datasets.

NCA-GA-SVM: A new two-level feature selection method based on neighborhood component analysis and genetic algorithm in hepatocellular carcinoma fatality prognosis.

Hepatocellular carcinoma (HCC) is one of the major challenges facing biomedical research. Despite the high lethality, methods to predict mortality for this type of aggressive malignant tumor are insufficient. Machine learning is recognized by many authors as a valuable, yet poorly studied tool in this field. Undoubtedly, searching for new feature selection methods is significant in building an effective machine-learning model. In this study, we propose the novel hybrid model using neighborhood components analysis, genetic algorithm and support vector machine classifier (NCA-GA-SVM). Because SVM works with default parameters characterized by low classification results, we decided to use GA for the proper optimization and feature selection. As reported in the available literature, NCA and GA obtain high classification results. Here, we decided to combine these approaches, building a two-level algorithm for HCC fatality prognosis. We used a well-known dataset collected from 165 patients at Coimbra's Hospital and University Center, Portugal. Our results revealed 96.36% classification accuracy and 95.52% F1-score. Additionally, we compared all data for these metrics published so far. We demonstrated that our algorithm achieved the highest accuracy and can be successfully applied for the assessment of hepatocellular carcinoma mortality in the future. Our findings bring methodological value for future HCC studies and emphasize the possibility of using machine-learning techniques to improve the quality of medical decisions.

A novel study for automatic two-class COVID-19 diagnosis (between COVID-19 and Healthy, Pneumonia) on X-ray images using texture analysis and 2-D/3-D convolutional neural networks.

The pandemic caused by the COVID-19 virus affects the world widely and heavily. When examining the CT, X-ray, and ultrasound images, radiologists must first determine whether there are signs of COVID-19 in the images. That is, COVID-19/Healthy detection is made. The second determination is the separation of pneumonia caused by the COVID-19 virus and pneumonia caused by a bacteria or virus other than COVID-19. This distinction is key in determining the treatment and isolation procedure to be applied to the patient. In this study, which aims to diagnose COVID-19 early using X-ray images, automatic two-class classification was carried out in four different titles: COVID-19/Healthy, COVID-19 Pneumonia/Bacterial Pneumonia, COVID-19 Pneumonia/Viral Pneumonia, and COVID-19 Pneumonia/Other Pneumonia. For this study, 3405 COVID-19, 2780 Bacterial Pneumonia, 1493 Viral Pneumonia, and 1989 Healthy images obtained by combining eight different data sets with open access were used. In the study, besides using the original X-ray images alone, classification results were obtained by accessing the images obtained using Local Binary Pattern (LBP) and Local Entropy (LE). The classification procedures were repeated for the images that were combined with the original images, LBP, and LE images in various combinations. 2-D CNN (Two-Dimensional Convolutional Neural Networks) and 3-D CNN (Three-Dimensional Convolutional Neural Networks) architectures were used as classifiers within the scope of the study. Mobilenetv2, Resnet101, and Googlenet architectures were used in the study as a 2-D CNN. A 24-layer 3-D CNN architecture has also been designed and used. Our study is the first to analyze the effect of diversification of input data type on classification results of 2-D/3-D CNN architectures. The results obtained within the scope of the study indicate that diversifying X-ray images with tissue analysis methods in the diagnosis of COVID-19 and including CNN input provides significant improvements in the results. Also, it is understood that the 3-D CNN architecture can be an important alternative to achieve a high classification result.