Effective Classification Model Research Articles

A Novel Approach for Predicting the Survival of Colorectal Cancer Patients Using Machine Learning Techniques and Advanced Parameter Optimization Methods

Background: Colorectal cancer is one of the most prevalent forms of cancer and is associated with a high mortality rate. Additionally, an increasing number of adults under 50 are being diagnosed with the disease. This underscores the importance of leveraging modern technologies, such as artificial intelligence, for early diagnosis and treatment support. Methods: Eight classifiers were utilized in this research: Random Forest, XGBoost, CatBoost, LightGBM, Gradient Boosting, Extra Trees, the k-nearest neighbor algorithm (KNN), and decision trees. These algorithms were optimized using the frameworks Optuna, RayTune, and HyperOpt. This study was conducted on a public dataset from Brazil, containing information on tens of thousands of patients. Results: The models developed in this study demonstrated high classification accuracy in predicting one-, three-, and five-year survival, as well as overall mortality and cancer-specific mortality. The CatBoost, LightGBM, Gradient Boosting, and Random Forest classifiers delivered the best performance, achieving an accuracy of approximately 80% across all the evaluated tasks. Conclusions: This research enabled the development of effective classification models that can be applied in clinical practice.

Vision Transformer Based Effective Model for Early Detection and Classification of Lung Cancer

Vision Transformer Based Effective Model for Early Detection and Classification of Lung Cancer

Feature Selection and Performance Evaluation of Buzzer Classification Model

In the rapidly evolving digital age, social media platforms have transformed into battleground for shaping public opinion. Among these platforms, X has been particularly susceptible to the phenomenon of 'buzzers', paid or coordinated actors who manipulate online discussions and influence public sentiment. This manipulation poses significant challenges for users, researchers, and policymakers alike, necessitating robust detection measures and strategic feature selection for accurate classification models. This research explores the utilization of various feature selection techniques to identify the most influential features among the 24 features employed in the classification modeling using Support Vector Machine. This study found that selecting 11 key features yields a remarkably effective classification model, achieving an impressive F1-score of 87.54 in distinguishing between buzzer and non-buzzer accounts. These results suggest that focusing on the relevant features can improve the accuracy and efficiency of buzzer detection models. By providing a more robust and adaptable solution to buzzer detection, our research has the potential to advance social media research and policy. This enabling researchers and policymakers to devise strategies aimed at mitigating misinformation dissemination and cultivating an environment of trust and integrity within social media platforms, thus fostering healthier online interactions and discourse.

Multi-stage ensemble with refinement for noisy labeled data classification

Deep neural networks (DNNs) have made remarkable progress in image classification. However, since DNNs can memorize all the label information in the training dataset due to their excellent feature learning ability, the generalization performance deteriorates when they are trained on the noisy labeled dataset that can be easily found in real-world problems. In this paper, we propose a multi-stage ensemble method with label refinement to build an effective classification model under noisy labels. The proposed method iteratively refines the dataset by re-labeling the samples at the end of each stage, which enables the models trained at each stage to learn different features. By integrating these models, the proposed multi-stage ensemble method exerts powerful generalization performance. Also, we suggest a novel dataset refinement method, demonstrating the effectiveness of a robust function in distinguishing corrupted samples. Experimental results on the benchmark and real-world datasets show that the proposed method outperforms the existing methods on the noisy labeled dataset classification.

CRE-TSCAE: A Novel Classification Model Based on Stacked Convolutional Autoencoder for Dual-Target RSVP-BCI Tasks.

The Rapid Serial Visual Presentation (RSVP) paradigm facilitates target identification in a rapid picture stream, which is applied extensively in military target surveillance and police monitoring. Most researchers concentrate on the single target RSVP-BCI whereas the study of dual-target is scarcely conducted, limiting RSVP application considerably. This paper proposed a novel classification model named Common Representation Extraction-Targeted Stacked Convolutional Autoencoder (CRE-TSCAE) to detect two targets with one nontarget in RSVP tasks. CRE generated a common representation for each target class to reduce variability from different trials of the same class and distinguish the difference between two targets better. TSCAE aimed to control uncertainty in the training process while requiring less target training data. The model learned a compact and discriminative feature through the training from several learning tasks so as to distinguish each class effectively. It was validated on the World Robot Contest 2021 and 2022 ERP datasets. Experimental results showed that CRE-TSCAE outperformed the state-of-the-art RSVP decoding algorithms and the Average ACC was 71.25%, improving 6.5% at least over the rest. It demonstrated that CRE-TSCAE showed a strong ability to extract discriminative latent features in detecting the differences among two targets with nontarget, which guaranteed increased classification accuracy. CRE-TSCAE provided an innovative and effective classification model for dual-target RSVP-BCI tasks and some insights into the neurophysiological distinction between different targets.

Rapid and nondestructive watermelon (Citrullus lanatus) seed viability detection based on visible near-infrared hyperspectral imaging technology and machine learning algorithms.

The improper storage of seeds can potentially compromise agricultural productivity, leading to reduced crop yields. Therefore, assessing seed viability before sowing is of paramount importance. Although numerous techniques exist for evaluating seed conditions, this research leveraged hyperspectral imaging (HSI) technology as an innovative, rapid, clean, and precise nondestructive testing method. The study aimed to determine the most effective classification model for watermelon seeds. Initially, purchased watermelon seeds were segregated into two groups: One underwent sterilization in a dehydrator machine at 40°C for 36h, whereas the other batch was stored under favorable conditions. Watermelon seeds' spectral images were captured using an HSI with a charge-coupled device camera ranging from 400 to 1000nm, and the segmented regions of all samples were measured. Preprocessing techniques and wavelength selection methods were applied to manage spectral data workload, followed by the implementation of a support vector machine (SVM) model. The initial hybrid-SVM model achieved a predictive accuracy rate of 100%, with a test set accuracy of 92.33%. Subsequently, an artificial bee colony (ABC) optimization was introduced to enhance model precision. The results indicated that, with kernel parameters (c, g) set at 13.17 and 0.01, respectively, and a runtime of 4.19328s, the training and evaluation of the dataset achieved an accuracy rate of 100%. Hence, it was practical to utilize HSI technology combined with the PCA-ABC-SVM model to detect different watermelon seeds. As a result, these findings introduce a novel technique for accurately forecasting seed viability, intended for use in agricultural industrial multispectral imaging. PRACTICAL APPLICATION: The traditional methods for determining the condition of seeds primarily emphasize aesthetics, rely on subjective assessment, are time-consuming, and require a lot of labor. On the other hand, HSI technology as green technology was employed to alleviate the aforementioned problems. This work significantly contributes to the field of industrial multispectral imaging by enhancing the capacity to discern various types of seeds and agricultural crop products.

Using Machine Learning to Evaluate the Value of Genetic Liabilities in the Classification of Hypertension within the UK Biobank.

Background and Objective: Hypertension increases the risk of cardiovascular diseases (CVD) such as stroke, heart attack, heart failure, and kidney disease, contributing to global disease burden and premature mortality. Previous studies have utilized statistical and machine learning techniques to develop hypertension prediction models. Only a few have included genetic liabilities and evaluated their predictive values. This study aimed to develop an effective hypertension classification model and investigate the potential influence of genetic liability for multiple risk factors linked to CVD on hypertension risk using the random forest and the neural network. Materials and Methods: The study involved 244,718 European participants, who were divided into training and testing sets. Genetic liabilities were constructed using genetic variants associated with CVD risk factors obtained from genome-wide association studies (GWAS). Various combinations of machine learning models before and after feature selection were tested to develop the best classification model. The models were evaluated using area under the curve (AUC), calibration, and net reclassification improvement in the testing set. Results: The models without genetic liabilities achieved AUCs of 0.70 and 0.72 using the random forest and the neural network methods, respectively. Adding genetic liabilities improved the AUC for the random forest but not for the neural network. The best classification model was achieved when feature selection and classification were performed using random forest (AUC = 0.71, Spiegelhalter z score = 0.10, p-value = 0.92, calibration slope = 0.99). This model included genetic liabilities for total cholesterol and low-density lipoprotein (LDL). Conclusions: The study highlighted that incorporating genetic liabilities for lipids in a machine learning model may provide incremental value for hypertension classification beyond baseline characteristics.

Cancer Classification from Gene Expression Using Ensemble Learning with an Influential Feature Selection Technique

Uncontrolled abnormal cell growth, known as cancer, may lead to tumors, immune system deterioration, and other fatal disability. Early cancer identification makes cancer treatment easier and increases the recovery rate, resulting in less mortality. Gene expression data play a crucial role in cancer classification at an early stage. Accurate cancer classification is a complex and challenging task due to the high-dimensional nature of the gene expression data relative to the small sample size. This research proposes using a dimensionality-reduction technique to address this limitation. Specifically, the mutual information (MI) technique is first utilized to select influential biomarker genes. Next, an ensemble learning model is applied to the reduced dataset using only the most influential features (genes) to develop an effective cancer classification model. The bagging method, where the base classifiers are Multilayer Perceptrons (MLPs), is chosen as an ensemble technique. The proposed cancer classification model, the MI-Bagging method, is applied to several benchmark gene expression datasets containing distinctive cancer classes. The cancer classification accuracy of the proposed model is compared with the relevant existing methods. The experimental results indicate that the proposed model outperforms the existing methods, and it is effective and competent for cancer classification despite the limited size of gene expression data with high dimensionality. The highest accuracy achieved by the proposed method demonstrates that the proposed emerging gene-expression-based cancer classifier has the potential to help in cancer treatment and lead to a higher cancer survival rate in the future.

Multi-label text classification on unbalanced Twitter with monolingual model and hyperparameter optimization for hate speech and abusive language detection

The increase in hate speech and abusive language on social media leads to uncomfortable interactions among users. Many datasets available publicly that address hate speech and abusive language are not balanced, particularly those from Indonesian Twitter. To develop a more effective classification model that also considers minority classes, we needed to optimize the hyperparameters of a monolingual model, use four different data preprocessing scenarios, and improve the treatment of slang words. We assessed the model's effectiveness by its accuracy, achieving 81.38%. This result came from optimizing hyperparameters, processing data without stemming and removing stop words, and enhancing the slang word data. The optimal hyperparameters were a learning rate of 4e-5, a batch size of 16, and a dropout rate of 0.1. However, using too much dropout can decrease the model’s performance and its ability to predict less common categories, such as physical- and gender-related hate speech.

Automatic Detection of Banana Maturity—Application of Image Recognition in Agricultural Production

With the development of machine vision technology, deep learning and image recognition technology has become a research focus for agricultural product non-destructive inspection. During the ripening process, banana appearance and nutrients clearly change, causing damage and unjustified economic loss. A high-efficiency banana ripeness recognition model was proposed based on a convolutional neural network and transfer learning. Banana photos at different ripening stages were collected as a dataset, and data augmentation was applied. Then, weights and parameters of four models trained on the original ImageNet dataset were loaded and fine-tuned to fit our banana dataset. To investigate the learning rate’s effect on model performance, fixed and updating learning rate strategies are analyzed. In addition, four CNN models, ResNet 34, ResNet 101, VGG 16, and VGG 19, are trained based on transfer learning. Results show that a slower learning rate causes the model to converge slowly, and the training loss function oscillates drastically. With different learning rate updating strategies, MultiStepLR performs the best and achieves a better accuracy of 98.8%. Among the four models, ResNet 101 performs the best with the highest accuracy of 99.2%. This research provides a direct effective model and reference for intelligent fruit classification.

Transcriptomic profile of premature ovarian insufficiency with RNA-sequencing.

This study aimed to explore the transcriptomic profile of premature ovarian insufficiency (POI) by investigating alterations in gene expression. A total of sixty-one women, comprising 31 individuals with POI in the POI group and 30 healthy women in the control group (HC group), aged between 24 and 40years, were recruited for this study. The transcriptomic profiles of peripheral blood samples from all study subjects were analyzed using RNA-sequencing. The results revealed 39 differentially expressed genes in individuals with POI compared to healthy controls, with 10 upregulated and 29 downregulated genes. Correlation analysis highlighted the relationship between the expression of SLC25A39, CNIH3, and PDZK1IP1 and hormone levels. Additionally, an effective classification model was developed using SLC25A39, CNIH3, PDZK1IP1, SHISA4, and LOC389834. Functional enrichment analysis demonstrated the involvement of these differentially expressed genes in the "haptoglobin-hemoglobin complex," while KEGG pathway analysis indicated their participation in the "Proteoglycans in cancer" pathway. The identified genes could play a crucial role in characterizing the genetic foundation of POI, potentially serving as valuable biomarkers for enhancing disease classification accuracy.

Optimized k-nearest neighbors for classification of prosthetic hand movements using electromyography signal

Electromyography (EMG) signals are essential, as they are used to measure muscular activity in different parts of the human body. The measurement and analysis of EMG signal lead to various applications of muscle disorders such as muscular dystrophy, myopathy, hand movements, etc. In this paper, an improved and effective hand movement classification model is developed for amputee subjects. It includes: (1) EMG feature extraction using Discrete Wavelet Transform (DWT), (2) EMG feature selection using binary Global Best Guided Gaussian Artificial Bee Colony (BGGABC), (3) Hand movements classification using Optimized k-nearest neighbors (OKNN) classifier. The EMG signal is taken from the DB3 of NinaPro dataset comprising 17 different prosthetic hand movements recorded from 11 amputee subjects. Thereafter, DWT is applied to decompose the EMG signal for extracting features. An improved wrapper-based feature selection technique (BGGABC) is used to select the optimal feature subset for effective classification. The two variants of KNN, i.e. Smallest Modified KNN and Largest Modified KNN are taken in which item strength to a class is optimized for efficient classification. The strength of an item to a class depends on distance and weight of an item to a class. Therefore, a multi-objective Non-dominated sorting genetic algorithm-II (NSGA-II) is used for optimizing these two contradictory parameters (distance and weight) simultaneously to have optimized variants, namely: Optimized Smallest KNN (OSKNN) and Optimized Largest KNN (OLKNN). Extensive results show that the proposed method OKNN achieved the highest classification accuracy of 93.07% (OLKNN) and 89.43% (OSKNN) compared with KNN variants and competitors.

Microbiome profiling and Co-metabolism pathway analysis in cervical cancer patients with acute radiation enteritis

BackgroundIntestinal bacteria significantly contribute to the metabolism of intestinal epithelial tissues. As the occurrence and development of radiation enteritis (RE) depend on the “co-metabolism” microenvironment formed by the host and intestinal microbiota, which involves complex influencing factors and strong correlations, ordinary techniques struggle to fully explain the underlying mechanisms. However, given that it is based on systems biology, metabolomics analysis is well-suited to address these issues. This study aimed to analyze the metabolomic changes in urine, serum, and fecal samples during volumetric modulated arc therapy (VMAT) for cervical cancer and screen for characteristic metabolites of severe acute radiation enteritis (SARE) and RE. MethodsWe enrolled 50 patients who received radiotherapy for cervical cancer. Urine, serum, and fecal samples of patients were collected at one day before radiotherapy and the second week, fourth week, and sixth week after the start of radiotherapy. Control group samples were collected during the baseline period. Differential metabolites were identified by metabolomics analysis; co-metabolic pathways were clarified. We used the mini-SOM library for incorporating characteristic metabolites, and established metabolite classification models for predicting SARE and RE. ResultsUrine and serum sample data showed remarkable clustering effect; metabolomics data of the fecal supernatant were evidently disturbed. Patient sample analyses during VMAT revealed the following. Urine samples: Downregulation of the pyrimidine and riboflavin metabolism pathways as well as initial upregulation followed by downregulation of arginine and proline metabolism pathways and the arginine biosynthesis pathway. Fecal samples: Upregulation of linoleic acid and phenylalanine metabolic pathways and initial downregulation followed by upregulation of arachidonic acid (AA) metabolic pathways. Serum samples: Initial upregulation followed by downregulation of the arginine biosynthesis pathway and downregulation of glutathione, AA, and arginine and proline metabolic pathways. ConclusionPatients with cervical cancer exhibited characteristic metabolic pathways and characteristic metabolites predicting RE and SARE were screened out. An effective RE mini-SOM classification model was successfully established.

Integrating Traditional Machine Learning and Deep Learning for Precision Screening of Anticancer Peptides: A Novel Approach for Efficient Drug Discovery.

The rapid and effective identification of anticancer peptides (ACPs) by computer technology provides a new perspective for cancer treatment. In the identification process of ACPs, accurate sequence encoding and effective classification models are crucial for predicting their biological activity. Traditional machine learning methods have been widely applied in sequence analysis, but deep learning provides a new approach to capture sequence complexity. In this study, a two-stage ACPs classification model was innovatively proposed. Three novel coding strategies were explored; two mainstream Natural Language Processing (NLP) models and 11 machine learning models were fused to identify ACPs, which significantly improved the prediction accuracy of ACPs. We analyzed the correlation between peptide chain amino acids and evaluated the relevant performance of the model by the ROC curve and t-SNE dimensionality reduction technique. The results indicated that the deep learning and machine learning fusion models of M3E-base and KNeighborsDist models, especially when considering the semantic information on amino acid sequences, achieved the highest average accuracy (AvgAcc) of 0.939, with an AUC value as high as 0.97. Then, in vitro cell experiments were used to verify that the two ACPs predicted by the model had antitumor efficacy. This study provides a convenient and effective method for screening ACPs. With further optimization and testing, these strategies have the potential to play an important role in drug discovery and design.

Research on lung sound classification model based on dual-channel CNN-LSTM algorithm

ulmonary diseases have a significant impact on human health and life safety, and abnormalities in the lungs are a direct response to lung diseases. Establishing an effective lung sound classification model that can assist in diagnosis is of great significance for electronic auscultation.In addressing the issue of lung sound signal classification, this study introduces a deep learning classification model based on a dual-channel CNN-LSTM algorithm. Initially, Mel-scale Frequency Cepstral Coefficients (MFCC) are employed for feature extraction from the dataset, transforming lung sound signals into Mel spectrograms. On this foundation, a dual-channel algorithm classification model is constructed, with parallel Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) modules. The CNN module is designed to capture spatial dimension features of the input data, while the LSTM module focuses on temporal dimension features. These two feature sets are fused together, enabling the model to classify lung sounds and thereby assisting in diagnosing pulmonary diseases for healthcare practitioners. This experiment used the ICBHI2017 Challenge Lungs dataset and obtained 5054 pieces of data through data augmentation and sampling techniques.The results show that the accuracy, recall, and F1 score of this model reach 99.01%, 99.13%, and 0.9915, respectively, significantly superior to other models, highlighting its practical application value.

Abstract 4878: Enhancing non-invasive detection of urothelial carcinoma through combined cytology and methylation profiling of urinary cell-free DNA

Abstract Detection of urothelial carcinoma (UC) faces a significant challenge in sensitivity due to the limitations of existing methodologies such as cytology. Identifying tumors through epigenetics pattern presents a promising advancement in diagnostic precision. This study aims to explore a non-invasive and more sensitive strategy by analyzing urinary cell-free DNA (ucfDNA) methylation profiles in conjunction with traditional urine cytology. The cohort for this study were collected from 20 patients with UC and 17 patients with benign or non-tumor lesions which histopathologic and cytopathologic results were reviewed by pathologist. Additional 16 cancer samples and 40 healthy donor samples were collected from independent research centers. We employed PredicineEPIC, a comprehensive whole-genome methylation assay, to investigate the DNA methylation characteristics in urine samples. The dataset was splited into 1:2 proportions for modeling and validation. The methylation signals were fed to three machine learning algorithms for determining the most effective classification model. Differentiating cancer and benign samples through urinary cytology has high specificity (1.0) but low sensitivity (0.3). Then we examined differentially methylated regions and implemented four most abnormally methylated regions as classify features. We found that the Gradient Boosting Machine (GBM) model achieved optimal performance in the validation set, with an AUC of 0.93, sensitivity of 0.81 (21/26), and specificity of 1.0 (36/36). The combination of methylation- and urine cytology-based models raised 2% prediction accuracy without reducing precision, suggesting that this assay could be applied to augment the cytology testing for UC detection. This study proposes a new and promising approach, leveraging cytology combined with epigenetic profiles of urinary cell-free DNA, for the precise diagnosis of urothelial carcinoma. Citation Format: Huan Zhao, Hang Dong, Haoran Tang, Zhihui Zhang, Na Wei, Jingjing Xu, Pan Du, Shidong Jia, Ting Xiao, Huiqin Guo. Enhancing non-invasive detection of urothelial carcinoma through combined cytology and methylation profiling of urinary cell-free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4878.

Utilizing ensemble learning in the classifications of ductile and brittle failure modes of UHPC strengthened RC members

This study aims to achieve the swift and precise classification of ductile and brittle failure modes in flexural reinforced concrete (RC) members, specifically those with tension sides strengthened by ultrahigh performance concrete (UHPC). Employing six ensemble learning techniques—Bagging, Random Forest, AdaBoost, Gradient Boosting, XGBoost, and LightGBM—the authors utilize a comprehensive dataset comprising 14 features, which include manually labeled failure modes obtain from load–deflection curves. The model training spans four scenarios, varying in the inclusion or exclusion of features describing the cross-sectional area of RC members and moment resistance. XGBoost emerges as the most effective classifier, achieving an impressive 84% accuracy with high confidence. Additionally, the study employs the Shapley Additive Explanation (SHAP) technique on the best-performing model to illuminate the significance and impacts of various features in UHPC-strengthened flexural members’ failure modes. Notably, moment resistance and UHPC tensile strength surface as the most influential factors in predicting failure modes. Increased rebar yield strength, UHPC compressive strength, UHPC reinforcement ratio, and steel fiber volume in UHPC contribute to enhanced ductility in flexural members, while heightened moment resistance and UHPC layer thickness, along with a robust RC-UHPC interface, tend to induce brittleness. The introduction of such an effective failure modes classification model, coupled with the model’s explainability, instills trust in its predictions and facilitates seamless integration into real-world applications, particularly in seismic areas. The model’s ability to operate without the need for pre-experimental tests marks a significant advancement in the field.

Machine learning tools in the tasks of selecting the determinants of socio-economic situation and development potential of Russian regions

Subject. The article addresses the selection of determinants that are significant for assessing the level of socio-economic situation and development potential of Russian regions. Objectives. The aim is to study machine learning algorithms for the selection of determinants – predictors of the level of socio-economic situation and development potential of Russian regions, to build models of classification of regions, according to the level of socio-economic situation, using various machine learning algorithms. Methods. To build classification models, I used data from the Federal State Statistics Service, the Institute of Scientific Communications, the RIA Novosti news agency, and the TAdviser Internet portal. Procedures for data classification, model parameter estimation, selection of significant determinants and visualization of results are performed, using the basic functions of the PyCaret library. Cohen's Kappa statistics and Matthews correlation coefficient were employed as priority metrics for evaluating the model productivity. The algorithms for selecting determinants are implemented in the Google Colab analytical environment. Results. I constructed multiclass classification models based on simple and ensemble machine learning algorithms. Simple classification algorithms, including logistic and ridge regression models, naive Bayesian algorithm, decision tree, support vector machine, and k-nearest neighbor methods are characterized by accuracy values at 77%, however, Cohen's Kappa statistics and the Matthews correlation coefficient only show a satisfactory relationship between the actual and predicted value of the region class. Ensemble algorithms, including random forest, gradient boosting and extreme gradient boosting, are characterized by a close relationship between the actual and forecast estimates of the classifier at a level of more than 70%. Conclusions. The random forest algorithm is recognized as the most effective classification model. The gross regional product and investments in fixed assets are informative determinants for measuring the socio-economic status.

Enhancing Aspect Category Detection Through Hybridised Contextualised Neural Language Models: A Case Study In Multi-Label Text Classification

Abstract Recently, the field of Natural Language Processing (NLP) has made significant progress with the evolution of Contextualised Neural Language Models (CNLMs) and the emergence of large LMs. Traditional and static language models exhibit limitations in tasks demanding contextual comprehension due to their reliance on fixed representations. CNLMs such as BERT and Semantic Folding aim to produce feature-rich representations by considering a broader linguistic context. In this paper, Deep Learning-based Aspect Category Detection approaches are introduced to perform text classification. The study extensively assesses classification model performance, emphasising enhanced representativeness and optimised feature extraction resolution using CNLMs and their hybridised variants. The effectiveness of the proposed approaches is evaluated on benchmark datasets of 4500 reviews from the laptop and restaurant domains. The results show that the proposed approaches using hybridised CNLMs outperform state-of-the-art methods with an f-score of 0.85 for the laptop and f-scores higher than 0.90 for the restaurant dataset. This study represents a pioneering work as one of the initial research efforts aiming to jointly evaluate the representation performance of CNLMs with different architectures to determine their classification capabilities. The findings indicate that the proposed approaches can enable the development of more effective classification models in various NLP tasks.

Automatic detection of small bowel lesions with different bleeding risks based on deep learning models.

Deep learning provides an efficient automatic image recognition method for small bowel (SB) capsule endoscopy (CE) that can assist physicians in diagnosis. However, the existing deep learning models present some unresolved challenges. To propose a novel and effective classification and detection model to automatically identify various SB lesions and their bleeding risks, and label the lesions accurately so as to enhance the diagnostic efficiency of physicians and the ability to identify high-risk bleeding groups. The proposed model represents a two-stage method that combined image classification with object detection. First, we utilized the improved ResNet-50 classification model to classify endoscopic images into SB lesion images, normal SB mucosa images, and invalid images. Then, the improved YOLO-V5 detection model was utilized to detect the type of lesion and its risk of bleeding, and the location of the lesion was marked. We constructed training and testing sets and compared model-assisted reading with physician reading. The accuracy of the model constructed in this study reached 98.96%, which was higher than the accuracy of other systems using only a single module. The sensitivity, specificity, and accuracy of the model-assisted reading detection of all images were 99.17%, 99.92%, and 99.86%, which were significantly higher than those of the endoscopists' diagnoses. The image processing time of the model was 48 ms/image, and the image processing time of the physicians was 0.40 ± 0.24 s/image (P < 0.001). The deep learning model of image classification combined with object detection exhibits a satisfactory diagnostic effect on a variety of SB lesions and their bleeding risks in CE images, which enhances the diagnostic efficiency of physicians and improves the ability of physicians to identify high-risk bleeding groups.