Handling class imbalance in COVID-19 chest X-ray images classification: Using SMOTE and weighted loss

Applied Soft Computing, Journal Year: 2022, Volume and Issue: 129, P. 109588 - 109588

Published: Aug. 29, 2022

Language: Английский

---

Toward Improving Breast Cancer Classification Using an Adaptive Voting Ensemble Learning Algorithm

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 12869 - 12882

Published: Jan. 1, 2024

Over the past decade, breast cancer has been most common type of in women. Different methods were proposed for detection. These mainly classify and categorize malignant Benign tumors. Machine learning is a practical approach classification. Data mining classification are effective to predict cancer. The optimum detecting Breast Cancer (BC) ensemble-based. ensemble involves using multiple ways find best possible solution. This study used Wisconsin Diagnostic (WBCD) dataset. We created voting classifier that combines four different machine models: Extra Trees Classifier (ETC), Light Gradient Boosting (LightGBM), Ridge (RC), Linear Discriminant Analysis (LDA). ELRL-E achieved an accuracy 97.6%, precision 96.4%, recall 100%, F1 score 98.1%. Various output evaluations evaluate performance efficiency model other classifiers. Overall, recommended strategy performed better. Results directly compared with individual recognized state-of-the-art primary objective this identify influential detection diagnosis terms AUC score.

Language: Английский

---

Novel hybrid of AOA-BSA with double adaptive and random spare for global optimization and engineering problems

Alexandria Engineering Journal, Journal Year: 2023, Volume and Issue: 73, P. 543 - 577

Published: May 11, 2023

Archimedes Optimization Algorithm (AOA) is a new physics-based optimizer that simulates principles. AOA has been used in variety of real-world applications because potential properties such as limited number control parameters, adaptability, and changing the set solutions to prevent being trapped local optima. Despite wide acceptance AOA, it some drawbacks, assumption individuals modify their locations depending on altered densities, volumes, accelerations. This causes various shortcomings stagnation into optimal regions, low diversity population, weakness exploitation phase, slow convergence curve. Thus, specific region conventional may be examined achieve balance between exploration capabilities AOA. The bird Swarm (BSA) an efficient strategy strong ability search process. In this study, hybrid called AOA-BSA proposed overcome limitations by replacing its phase with BSA one. Moreover, transition operator have high exploitation. To test examine performance, first experimental series, 29 unconstrained functions from CEC2017 whereas series second experiments use seven constrained engineering problems AOA-BSA's handling issues. performance suggested algorithm compared 10 optimizers. These are original algorithms 8 other algorithms. experiment's results show effectiveness optimizing suite. AOABSA outperforms metaheuristic across 16 functions. statically validated using Wilcoxon Rank sum. shows superior capability. due added power integration not only seen faster achieved AOABSA, but also found For further validation extensive statistical analysis performed during process recording ratios problems, achieves competitive curve reaches lowest values problem. It minimum standard deviation which indicates robustness solving these problems. Also, obtained counterparts regarding problem variables behavior best values.

Language: Английский

---

Classification and Localization of Multi-Type Abnormalities on Chest X-Rays Images

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 83264 - 83277

Published: Jan. 1, 2023

Chest X-ray images are among the most common diagnostic tools for detecting and managing bronchopneumonia lung abnormalities, such as those caused by COVID-19. However, interpreting these requires significant expertise, misinterpretations can result in false negatives or positives. Deep learning techniques have recently been highly effective analyzing medical images, including chest X-rays. In this study, we propose two deep approaches to classify localize different COVID-19, on X-rays, which include multi-classification object detection models that identify presence of disease other abnormalities. The proposed trained a large dataset from sick people (including COVID-19 patients) validated an independent test set. Compared single models, paper presents ensemble combining multiple detect abnormalities images. Our results demonstrate method achieved promising both localization compared state-of-the-art methodologies. methods potential assist radiologists diagnosis provide more accurate efficient interpretation, thereby improving patient outcomes reducing burden healthcare systems.

Language: Английский

---

Improved classification of colorectal polyps on histopathological images with ensemble learning and stain normalization

Computer Methods and Programs in Biomedicine, Journal Year: 2023, Volume and Issue: 232, P. 107441 - 107441

Published: Feb. 24, 2023

Language: Английский

---

Deep Learning-Based Model Using DensNet201 for Mobile User Interface Evaluation

International Journal of Human-Computer Interaction, Journal Year: 2023, Volume and Issue: 39(9), P. 1981 - 1994

Published: Feb. 12, 2023

Human-centered AI plays a vital role in ensuring that human capabilities and ideas are tailored to meet efficiently the data requirements. The main idea is focusing on making machines learn from behavior many fields including e-learning, mobile computing, e-health. In this context, sales of devices rising every day Mobile User Interfaces (MUI) for smartphones tablets attracting greater attention. way, there widely development tools new services. Moreover, useful apps help users their daily living such as health, entertainment, games, social networking, weather, logistics transport. user interfaces have become necessity user’s satisfaction. evaluation fundamental dimension success apps. Generally, two classes interface methods: manual automatic. first category conducted by or experts evaluate visual design quality MUIs. Nevertheless, it more time-consuming task. second used an automatic tool, require preconfiguration source code. However, configuration difficult task non-programmer evaluators. To address issue, we propose method based analysis graphical MUI screenshot without using code participation. proposed combines Densnet201 architecture K-Nearest Neighbours (KNN) classifier assess First, apply Borderline-SMOTE obtain balanced dataset. Then, GoogleNet extract automatically features MUI. Finally, KNN classify MUIs good bad. We approach publicly available large-scale datasets. obtained results very promising shows efficiency model with average 93% accuracy. This implemented application designers aims improving fact, can decrease misunderstanding needs improve usability order reach

Language: Английский

---

Tunicate swarm algorithm with deep convolutional neural network-driven colorectal cancer classification from histopathological imaging data

Electronic Research Archive, Journal Year: 2023, Volume and Issue: 31(5), P. 2793 - 2812

Published: Jan. 1, 2023

<abstract> <p>Colorectal cancer (CRC) is one of the most popular cancers among both men and women, with increasing incidence. The enhanced analytical load data from pathology laboratory, integrated described intra- inter-variabilities through calculation biomarkers, has prompted quest for robust machine-based approaches in combination routine practice. In histopathology, deep learning (DL) techniques have been applied at large due to their potential supporting analysis forecasting medically appropriate molecular phenotypes microsatellite instability. Considering this background, current research work presents a metaheuristics technique convolutional neural network-based colorectal classification based on histopathological imaging (MDCNN-C3HI). presented MDCNN-C3HI majorly examines images (CRC). At initial stage, applies bilateral filtering approach get rid noise. Then, proposed uses an capsule network Adam optimizer extraction feature vectors. For CRC classification, DL modified classifier, whereas tunicate swarm algorithm used fine-tune its hyperparameters. To demonstrate performance wide range experiments was conducted. outcomes extensive experimentation procedure confirmed superior over other existing techniques, achieving maximum accuracy 99.45%, sensitivity 99.45% specificity 99.45%.</p> </abstract>

Language: Английский

---

An effective colorectal polyp classification for histopathological images based on supervised contrastive learning

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 172, P. 108267 - 108267

Published: March 8, 2024

Early detection of colon adenomatous polyps is pivotal in reducing cancer risk. In this context, accurately distinguishing between polyp subtypes, especially tubular and tubulovillous, from hyperplastic variants crucial. This study introduces a cutting-edge computer-aided diagnosis system optimized for task. Our employs advanced Supervised Contrastive learning to ensure precise classification histopathology images. Significantly, we have integrated the Big Transfer model, which has gained prominence its exemplary adaptability visual tasks medical imaging. novel approach discerns in-class out-of-class images, thereby elevating discriminatory power subtypes. We validated our using two datasets: specially curated one publicly accessible UniToPatho dataset. The results reveal that model markedly surpasses traditional deep convolutional neural networks, registering accuracies 87.1% 70.3% custom datasets, respectively. Such emphasize transformative potential endeavors.

Language: Английский

---

MTU: A multi-tasking U-net with hybrid convolutional learning and attention modules for cancer classification and gland Segmentation in Colon Histopathological Images

Computers in Biology and Medicine, Journal Year: 2022, Volume and Issue: 150, P. 106095 - 106095

Published: Sept. 21, 2022

Language: Английский

---

Origin-destination prediction from road average speed data using GraphResLSTM model

PeerJ Computer Science, Journal Year: 2025, Volume and Issue: 11, P. e2709 - e2709

Published: Feb. 13, 2025

With the increasing demand for traffic management and resource allocation in Intelligent Transportation Systems (ITS), accurate origin-destination (OD) prediction has become crucial. This article presents a novel integrated framework, effectively merging distinctive capabilities of graph convolutional network (GCN), residual neural (ResNet), long short-term memory (LSTM), hereby designated as GraphResLSTM. GraphResLSTM leverages road average speed data OD prediction. Contrary to traditional reliance on flow data, provides richer informational dimensions, reflecting not only vehicle volume but also indirectly indicating congestion levels. We use real-world generate through simulations Simulation Urban Mobility (SUMO), thereby avoiding influence external factors such weather. To enhance training efficiency, we employ method combining entropy weight with Technique Order Preference by Similarity Ideal Solution (TOPSIS) key segment selection. Using this generated dataset, carefully designed comparative experiments are conducted compare various different models types. The results clearly demonstrate that both model markedly outperform alternative types

Language: Английский

---