The paper presents an efficient framework for COVID-19 detection using Vision Transformers (ViT) and Explainable AI (XAI) techniques. The key highlights are:
Image Preprocessing: The framework employs contrast limited adaptive histogram equalization (CLAHE) and the Ben Graham method to enhance the quality of the input X-ray and CT scan images, improving the performance of the predictive models.
Data Augmentation: The paper utilizes various image augmentation techniques such as Gaussian blur, random rotation, zooming, and flipping to increase the diversity of the training data and improve the model's generalization.
Compact Convolutional Transformers (CCT): The authors propose a CCT model that combines convolutional blocks and transformer encoders to effectively capture spatial relationships and global patterns in the input images. CCT outperforms the standard ViT approach in terms of accuracy and efficiency.
Explainable AI (XAI): The paper employs Gradient-weighted Class Activation Mapping (Grad-CAM) to generate heatmaps that highlight the regions in the input images that are most important for the COVID-19 classification, providing interpretability and insights into the model's decision-making process.
Evaluation: The proposed framework is evaluated on the COVID-19 Radiography Database, achieving a training accuracy of 97% and a validation accuracy of 94.6%. The model's performance is further analyzed using various metrics such as precision, recall, F1-score, and confusion matrix.
The comprehensive approach presented in this paper, combining advanced image processing, data augmentation, transformer-based architecture, and XAI techniques, demonstrates a robust and interpretable solution for COVID-19 detection from medical imaging data.
To Another Language
from source content
arxiv.org
Ключові висновки, отримані з
by Pangoth Sant... о arxiv.org 05-07-2024
https://arxiv.org/pdf/2307.16033.pdfГлибші Запити