GastroVRG: Enhancing early screening in gastrointestinal health via advanced transfer features

Intelligent Systems with Applications, Journal Year: 2024, Volume and Issue: 23, P. 200399 - 200399

Published: June 20, 2024

The accurate classification of endoscopic images is a challenging yet critical task in medical diagnostics, which directly affects the treatment and management Gastrointestinal diseases. Misclassification can lead to incorrect plans, adversely affecting patient outcomes. To address this challenge, our research aimed develop reliable computational model improve accuracy classifying conditions esophagitis polyps. We focused on subset Kvasir v1 secondary dataset, comprising 2000 evenly distributed across two classes: polyp. goal was leverage strengths both Machine Learning(ML) Deep Learning(DL) create that not only predicts with high but also integrates seamlessly into clinical workflows. end, we introduced novel VRG-based ensemble image feature extraction technique, combining powers VGG, RF, GB models synthesize robust set conducive high-precision classification. approach demonstrated best-in-class performance achieving an outstanding 99.73% detecting practical implications these results are substantial, indicating method significantly diagnostic real-world settings, reduce rate misdiagnosis, contribute efficient effective patients, ultimately enhancing quality healthcare services. With successful application proposed controlled future work involves deploying environments expanding its broader spectrum multi-class datasets.

Language: Английский

---

Recent Advancements in Localization Technologies for Wireless Capsule Endoscopy: A Technical Review

Sensors, Journal Year: 2025, Volume and Issue: 25(1), P. 253 - 253

Published: Jan. 4, 2025

Conventional endoscopy is limited in its ability to examine the small bowel and perform long-term monitoring due risk of infection tissue perforation. Wireless Capsule Endoscopy (WCE) a painless non-invasive method examining body’s internal organs using camera that swallowed like pill. The existing active locomotion technologies do not have practical localization system control capsule’s movement within body. A robust essential for safely guiding WCE device through complex gastrointestinal (GI) tract. Moreover, having access trajectory data highly desirable drug delivery surgery, as well creating accurate user profiles diagnosis future reference. Therefore, robust, real-time, imperative advance field make it clinical trials. In this work, we identified salient features different techniques categorized studies comprehensive tables. This study self-contained offers overview emerging based on magnetic field, radio frequency (RF), video, hybrid methods. summary at end each provided point out potential gaps give directions research. main work present an in-depth review most recent published past five years. will assist researchers comprehending current pinpointing areas further investigation. can be significant reference guide research localization.

Language: Английский

---

Wireless Capsule Endoscopy Image Classification: An Explainable AI Approach

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 105262 - 105280

Published: Jan. 1, 2023

Deep Learning has contributed significantly to the advances made in fields of Medical Imaging and Computer Aided Diagnosis (CAD). Although a variety (DL) models exist for purposes image classification medical domain, more analysis needs be conducted on their decision-making processes. For this reason, several novel Explainable AI (XAI) techniques have been proposed recent years better understand DL models. Currently, professionals rely visual inspections diagnose potential diseases endoscopic imaging preliminary stages. However, we believe that use automated systems can enhance both efficiency such diagnoses. The aim study is increase reliability model predictions within field by implementing transfer learning balanced subset Kvasir-capsule, Wireless Capsule Endoscopy dataset. This includes top 9 classes dataset training testing. results obtained were an F1-score 97%±1% Vision Transformer model, although other as MobileNetv3Large ResNet152v2 also able achieve F1-scores over 90%. These are currently highest-reported metrics data, improving upon prior studies done same heatmaps XAI techniques, including GradCAM, GradCAM++, LayersCAM, LIME, SHAP presented form evaluated according highlighted regions importance. effort decisions top-performing look beyond black-box nature.

Language: Английский

---

A new capsule-intestine model for the capsule robot self-propelling in the lower gastrointestinal tract

European Journal of Mechanics - A/Solids, Journal Year: 2024, Volume and Issue: 105, P. 105233 - 105233

Published: Jan. 10, 2024

Circular and haustral folds in the lower gastrointestinal tract (small large intestines) are major obstructions impeding locomotion of capsule robots for endoscopic diagnosis. Understanding interactions between these is critical design control to reach areas clinical interest. This paper proposes a new mathematical model capsule-intestine interaction based on our previous work (Yan et al. 2022) by introducing capsule's rotation during fold crossing. The resisting force predicted more consistent with finite element experimental results compared model. It found that obstructive effect stronger higher thinner stiffer intestine. For robot, which actuated periodically driven inner mass, excitation required overcome larger force. Moreover, bifurcation analysis reveals small always incurs simple period-1 motion while may result various complex dynamics before findings this help robotics engineers evaluate their designs terms propulsion understand tract.

Language: Английский

---

An effective object detection and tracking using automated image annotation with inception based faster R-CNN model

International Journal of Cognitive Computing in Engineering, Journal Year: 2024, Volume and Issue: 5, P. 343 - 356

Published: Jan. 1, 2024

The present study advances object detection and tracking techniques by proposing a novel model combining Automated Image Annotation with Inception v2-based Faster RCNN (AIA-IFRCNN). research methodology utilizes the DCF-CSRT for image annotation, detection, inception v2 feature extraction, followed softmax layer classification. proposed AIA-IFRCNN is evaluated on three benchmark datasets: Bird (Dataset 1), UCSDped2 2), Under Water 3), to determine prediction accuracy, annotation time, Center Location Error (CLE), Overlap Rate (OR). experimental results indicate that outperformed existing models regarding accuracy performance. Notably, it achieved maximum of 95.62 % Dataset 1, outperforming other models. Additionally, minimum average CLE values 4.16, 5.78, 3.54, higher overlap rates 0.92, 0.90, 0.94 respective datasets (1, 2 3). Hence, this work using essential improving system efficiency fostering innovation in field computer vision tracking.

Language: Английский

---

Study the Capacity of Deep Learning Techniques Information Generalization Using Capsule Endoscopic Images

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 380 - 394

Published: Jan. 1, 2025

Language: Английский

---

The analysis of motion recognition model for badminton player movements using machine learning

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 30, 2025

Language: Английский

---

Federated Deep Learning for Wireless Capsule Endoscopy Analysis: Enabling Collaboration Across Multiple Data Centers for Robust Learning of Diverse Pathologies

Future Generation Computer Systems, Journal Year: 2023, Volume and Issue: 152, P. 361 - 371

Published: Oct. 7, 2023

Language: Английский

---

Human-like evaluation by facial attractiveness intelligent machine

International Journal of Cognitive Computing in Engineering, Journal Year: 2023, Volume and Issue: 4, P. 160 - 169

Published: April 12, 2023

Facial attractiveness is an important factor in social interactions and has been widely studied psychology neuroscience. This paper presents a novel approach to the problem of predicting facial using machine learning computer vision techniques. Our main objective investigate whether intelligent can learn accurately predict based on rules features. To achieve this, we collected datasets images corresponding rankings for women. We then utilized various methods, including k-nearest neighbors (KNN) support vector regression (SVR), train predictor model that learned from these provide human-like assessment attractiveness. The used feature parameters, such as symmetry proportion, input determine ranking output. evaluated performance our trained several metrics, coefficient determination (R2), root-mean-square error (RMSE), mean absolute percentage (MAPE). best was achieved KNN algorithm during testing phase, with R2=0.9902, RMSE=0.0056, MAPE=0.0856. It indicated significant improvement accuracy prediction compared previous studies. results demonstrate features, providing promising In comparison studies this area, shows accuracy, correlation higher than human ratings. work implications fields psychology, neuroscience, science, it provides new perspective concept its quantification learning.

Language: Английский

---

Development and validation of a deep learning system for detection of small bowel pathologies in capsule endoscopy: a pilot study in a Singapore institution

Singapore Medical Journal, Journal Year: 2024, Volume and Issue: 65(3), P. 133 - 140

Published: March 1, 2024

Abstract Introduction: Deep learning models can assess the quality of images and discriminate among abnormalities in small bowel capsule endoscopy (CE), reducing fatigue time needed for diagnosis. They serve as a decision support system, partially automating diagnosis process by providing probability predictions abnormalities. Methods: We demonstrated use deep CE image analysis, specifically piloting preparation model (BPM) an abnormality detection (ADM) to determine frame-level view presence abnormal findings, respectively. used convolutional neural network-based pretrained on large-scale open-domain data extract spatial features that were then dense feed-forward network classifier. combined open-source Kvasir-Capsule dataset ( n = 43) locally collected 29). Results: Model performance was compared using averaged five-fold two-fold cross-validation BPMs ADMs, The best BPM based pre-trained ResNet50 architecture had area under receiver operating characteristic precision-recall curves 0.969±0.008 0.843±0.041, ADM model, also ResNet50, top-1 top-2 accuracies 84.03±0.051 94.78±0.028, could approximately 200–250 per second showed good discrimination time-critical such bleeding. Conclusion: Our pilot potential improve workflows. To our knowledge, approach is unique Singapore context. value work be further evaluated pragmatic manner sensitive existing clinician workflow resource constraints.

Language: Английский

---