Time-based self-supervised learning for Wireless Capsule Endoscopy

Computers in Biology and Medicine, Journal Year: 2022, Volume and Issue: 146, P. 105631 - 105631

Published: May 24, 2022

State-of-the-art machine learning models, and especially deep ones, are significantly data-hungry; they require vast amounts of manually labeled samples to function correctly. However, in most medical imaging fields, obtaining said data can be challenging. Not only the volume is a problem, but also imbalances within its classes; it common have many more images healthy patients than those with pathology. Computer-aided diagnostic systems suffer from these issues, usually over-designing their models perform accurately. This work proposes using self-supervised for wireless endoscopy videos by introducing custom-tailored method that does not initially need labels or appropriate balance. We prove inferred inherent structure learned our method, extracted temporal axis, improves detection rate on several domain-specific applications even under severe imbalance.

Language: Английский

---

EnsemDeepCADx: Empowering Colorectal Cancer Diagnosis with Mixed-Dataset Features and Ensemble Fusion CNNs on Evidence-Based CKHK-22 Dataset

Bioengineering, Journal Year: 2023, Volume and Issue: 10(6), P. 738 - 738

Published: June 19, 2023

Colorectal cancer is associated with a high mortality rate and significant patient risk. Images obtained during colonoscopy are used to make diagnosis, highlighting the importance of timely diagnosis treatment. Using techniques deep learning could enhance diagnostic accuracy existing systems. most advanced techniques, brand-new EnsemDeepCADx system for accurate colorectal has been developed. The optimal achieved by combining Convolutional Neural Networks (CNNs) transfer via bidirectional long short-term memory (BILSTM) support vector machines (SVM). Four pre-trained CNN models comprise ADaDR-22, ADaR-22, DaRD-22 ensemble CNNs: AlexNet, DarkNet-19, DenseNet-201, ResNet-50. In each its stages, CADx thoroughly evaluated. From CKHK-22 mixed dataset, colour, greyscale, local binary pattern (LBP) image datasets features utilised. second stage, returned compared new feature fusion dataset using three distinct ensembles. Next, they incorporate CNNs SVM-based comparing raw datasets. final stage learning, BILSTM SVM combined ensemble. testing DarD-22 on original, grey, LBP, was (95.96%, 88.79%, 73.54%, 97.89%). Comparing outputs all four those at enables attain highest level accuracy.

Language: Английский

---

An advanced diagnostic ColoRectalCADx utilises CNN and unsupervised visual explanations to discover malignancies

Neural Computing and Applications, Journal Year: 2023, Volume and Issue: 35(28), P. 20631 - 20662

Published: July 22, 2023

Language: Английский

---

ColoRectalCADx: Expeditious Recognition of Colorectal Cancer with Integrated Convolutional Neural Networks and Visual Explanations Using Mixed Dataset Evidence

Computational and Mathematical Methods in Medicine, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 27

Published: Nov. 10, 2022

Colorectal cancer typically affects the gastrointestinal tract within human body. Colonoscopy is one of most accurate methods detecting cancer. The current system facilitates identification by computer-assisted diagnosis (CADx) systems with a limited number deep learning methods. It does not imply depiction mixed datasets for functioning system. proposed system, called ColoRectalCADx, supported (DL) models suitable research. CADx comprises five stages: convolutional neural networks (CNN), support vector machine (SVM), long short-term memory (LSTM), visual explanation such as gradient-weighted class activation mapping (Grad-CAM), and semantic segmentation phases. Here, key components are equipped 9 individual 12 integrated CNNs, implying that consists mainly investigational experiments total 21 CNNs. In subsequent phase, has combination CNNs concatenated transfer functions associated SVM classification. Additional classification applied to ensure effective results from CNN LSTM. made up CVC Clinic DB, Kvasir2, Hyper Kvasir input dataset. After LSTM, in advanced stage, malignancies detected using better polyp recognition technique Grad-CAM U-Net. have been stored on Google Cloud record retention. these experiments, among all DenseNet-201 (87.1% training 84.7% testing accuracies) ADaDR-22 (84.61% 82.17% were efficient detection CNN+LSTM model. ColoRectalCADx accurately identifies through DesnseNet-201 ADaDR-22. Grad-CAM's explanations, displays precise visualization polyps, U-Net provides malignant polyps.

Language: Английский

---

Medical image classification using a combination of features from convolutional neural networks

Multimedia Tools and Applications, Journal Year: 2022, Volume and Issue: 82(13), P. 19299 - 19322

Published: Nov. 15, 2022

Language: Английский

---