Breast Cancer Detection on Histopathological Images Using a Composite Dilated Backbone Network

Computational Intelligence and Neuroscience, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 10

Published: July 6, 2022

Breast cancer is a lethal illness that has high mortality rate. In treatment, the accuracy of diagnosis crucial. Machine learning and deep may be beneficial to doctors. The proposed backbone network critical for present performance CNN-based detectors. Integrating dilated convolution, ResNet, Alexnet increases detection performance. composite (CDBN) an innovative method integrating many identical backbones into single robust backbone. Hence, CDBN uses lead feature maps identify objects. It feeds high-level output features from previous next in stepwise way. We show most contemporary detectors can easily include improve achieved mAP improvements ranging 1.5 3.0 percent on breast histopathological image classification (BreakHis) dataset. Experiments have also shown instance segmentation improved. BreakHis dataset, enhances baseline detector cascade mask R-CNN (mAP = 53.3). does not need pretraining. creates traits by combining low-level elements. This made up several are linked together. considers CDBN.

Language: Английский

---

Data augmentation for medical imaging: A systematic literature review

Computers in Biology and Medicine, Journal Year: 2022, Volume and Issue: 152, P. 106391 - 106391

Published: Dec. 9, 2022

Language: Английский

---

Pooling in convolutional neural networks for medical image analysis: a survey and an empirical study

Neural Computing and Applications, Journal Year: 2022, Volume and Issue: 34(7), P. 5321 - 5347

Published: Feb. 1, 2022

Language: Английский

---

Explainable Artificial Intelligence for Human Decision Support System in the Medical Domain

Machine Learning and Knowledge Extraction, Journal Year: 2021, Volume and Issue: 3(3), P. 740 - 770

Published: Sept. 19, 2021

In this paper, we present the potential of Explainable Artificial Intelligence methods for decision support in medical image analysis scenarios. Using three types explainable applied to same data set, aimed improve comprehensibility decisions provided by Convolutional Neural Network (CNN). vivo gastral images obtained a video capsule endoscopy (VCE) were subject visual explanations, with goal increasing health professionals’ trust black-box predictions. We implemented two post hoc interpretable machine learning methods, called Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP), an alternative explanation approach, Contextual Importance Utility (CIU) method. The produced explanations assessed human evaluation. conducted user studies based on LIME, SHAP CIU. Users from different non-medical backgrounds carried out series tests web-based survey setting stated their experience understanding given explanations. Three groups (n = 20, 20) distinct forms quantitatively analyzed. found that, as hypothesized, CIU-explainable method performed better than both LIME terms improving decision-making being more transparent thus understandable users. Additionally, CIU outperformed generating rapidly. Our findings suggest that there are notable differences between various settings. line future improvements implementation, can be generalized sets provide effective experts.

Language: Английский

---

MIDCAN: A multiple input deep convolutional attention network for Covid-19 diagnosis based on chest CT and chest X-ray

Pattern Recognition Letters, Journal Year: 2021, Volume and Issue: 150, P. 8 - 16

Published: July 14, 2021

Language: Английский

---

Hybrid quantum-classical convolutional neural network model for COVID-19 prediction using chest X-ray images

Journal of Computational Design and Engineering, Journal Year: 2022, Volume and Issue: 9(2), P. 343 - 363

Published: Jan. 10, 2022

Despite the great efforts to find an effective way for COVID-19 prediction, virus nature and mutation represent a critical challenge diagnose covered cases. However, developing model predict via Chest X-Ray (CXR) images with accurate performance is necessary help in early diagnosis. In this paper, hybrid quantum-classical convolutional Neural Networks (HQCNN) used random quantum circuits (RQCs) as base detect patients CXR images. A collection of 6952 images, including 1161 COVID-19, 1575 normal, 5216 pneumonia were dataset work. The proposed HQCNN achieved higher accuracy 98.4\% sensitivity 99.3\% on first Besides, it obtained 99\% 99.7\% second Also, accuracy, 88.6\%, 88.7\%, respectively, third multi-class Furthermore, outperforms various models balanced precision, F1-measure, AUC-ROC score. experimental results are by prove its ability predicting positive

Language: Английский

---

Explainable Artificial Intelligence Methods in Combating Pandemics: A Systematic Review

IEEE Reviews in Biomedical Engineering, Journal Year: 2022, Volume and Issue: 16, P. 5 - 21

Published: June 23, 2022

Despite the myriad peer-reviewed papers demonstrating novel Artificial Intelligence (AI)-based solutions to COVID-19 challenges during pandemic, few have made a significant clinical impact, especially in diagnosis and disease precision staging. One major cause for such low impact is lack of model transparency, significantly limiting AI adoption real practice. To solve this problem, models need be explained users. Thus, we conducted comprehensive study Explainable (XAI) using PRISMA technology. Our findings suggest that XAI can improve performance, instill trust users, assist users decision-making. In systematic review, introduce common techniques their utility with specific examples application. We discuss evaluation results because it an important step maximizing value AI-based decision support systems. Additionally, present traditional, modern, advanced demonstrate evolution techniques. Finally, provide best practice guideline developers refer experimentation. also offer potential This hopefully, promote biomedicine healthcare.

Language: Английский

---

Automated detection and forecasting of COVID-19 using deep learning techniques: A review

Neurocomputing, Journal Year: 2024, Volume and Issue: 577, P. 127317 - 127317

Published: Jan. 26, 2024

Language: Английский

---

ADVIAN: Alzheimer's Disease VGG-Inspired Attention Network Based on Convolutional Block Attention Module and Multiple Way Data Augmentation

Frontiers in Aging Neuroscience, Journal Year: 2021, Volume and Issue: 13

Published: June 18, 2021

Aim: Alzheimer's disease is a neurodegenerative that causes 60–70% of all cases dementia. This study to provide novel method can identify AD more accurately. Methods: We first propose VGG-inspired network (VIN) as the backbone and investigate use attention mechanisms. proposed an Disease VGG-Inspired Attention Network (ADVIAN), where we integrate convolutional block modules on VIN backbone. Also, 18-way data augmentation avoid overfitting. Ten runs 10-fold cross-validation are carried out report unbiased performance. Results: The sensitivity specificity reach 97.65 ± 1.36 97.86 1.55, respectively. Its precision accuracy 97.87 1.53 97.76 1.13, F1 score, MCC, FMI obtained 97.75 95.53 2.27, AUC 0.9852. Conclusion: ADVIAN gives better results than 11 state-of-the-art methods. Besides, experimental demonstrate effectiveness augmentation.

Language: Английский

---

Explainable Deep Learning-Based Approach for Multilabel Classification of Electrocardiogram

IEEE Transactions on Engineering Management, Journal Year: 2021, Volume and Issue: 70(8), P. 2787 - 2799

Published: Sept. 14, 2021

Recently computer-aided diagnosis methods have been widely adopted to aid doctors in disease making their decisions more reliable and error-free. Electrocardiogram (ECG) is the most commonly used, noninvasive diagnostic tool for investigating various cardiovascular diseases. In real life, patients suffer from than one heart at a time. So any practical automated system should identify multiple diseases present single ECG signal. this article, we propose novel deep learning-based method multilabel classification of signals. The proposed can accurately up two labels an signal pertaining eight rhythm or morphological abnormalities also normal condition. Also, black-box nature learning models prevents them being applied high-risk like diagnosis. establish explainable artificial intelligence (XAI) framework using class activation maps obtained Grad-CAM technique. method, train convolutional neural network (CNN) with constructed matrices. With experiments conducted, that training CNN by taking only label each data point enough learn features information it (multiple same time). During classification, apply thresholding on output probabilities softmax layer our CNN, obtain signals.We trained model 6311 records tested 280 records. testing, achieved subset accuracy 96.2% hamming loss 0.037 precision 0.986 recall 0.949 F1-score 0.967. Considering fact has performed very well all metrics be directly used as

Language: Английский

---