Application of CBAM (Convolutional Block Attention Module) for CT Scan Image Classification

With the increasing demand for accurate medical image classification, advanced deep learning techniques are becoming indispensable. This study aims to delve into the potential of the Convolutional Block Attention Module (CBAM) in significantly improving the classification accuracy of CT scan images. Unlike traditional convolutional neural networks (CNNs), CBAM introduces a mechanism to emphasize important features and spatial locations within the image. By dynamically recalibrating feature maps, CBAM can effectively highlight crucial information, thereby promising to achieve superior performance compared to conventional CNN architectures. This research seeks to demonstrate how integrating CBAM into the classification pipeline can lead to more precise and reliable diagnoses in medical imaging, potentially revolutionizing the field by harnessing the power of attention mechanisms in deep learning. The main goal of our experiment is to conduct a comparative study between configurable CNN and configurable CNN enhanced with CBAM.

Vision transformer and CNN models for X-RAY image classification

Recently, significant advancements have been made in the image classification of pneumonia, COVID-19, and healthy cases by utilizing Convolutional Neural Networks (CNNs) and Vision Transformer (ViT) models. While CNNs have been extensively employed in the literature for image classification tasks, ViT models have recently emerged and showcased notable improvements in classification accuracy. It is crucial to compare the effectiveness of these models and optimize their utilization in clinical settings to facilitate early detection and diagnosis of pulmonary diseases. This study presents a comparative analysis of CNNs and ViT models for X-ray image classification of COVID-19, pneumonia, and healthy cases. The investigation primarily focuses on evaluating their performance. The findings contribute to enhancing the understanding of these models and provide valuable insights to improve the accuracy and efficiency of pulmonary disease detection. The study emphasizes the potential of both CNNs and ViT models in assisting healthcare professionals in the diagnosis and treatment of pulmonary diseases. The main goal of our experiment is to train both CNN models VGG16, ResNet-50 and the vision transformer, the VIT models reached an accuracy of 85 % these results are taken with imbalanced data and without preprocessing for our data.

Classification des pathologies pulmonaires COVID-19 à l'aide d'un apprentissage profond à partir d'images échographiques

In recent years, many classification algorithms have been implemented to distinguishing between different lung diseases. Machine learning techniques have been widely used for lung disease classification and have particularly focused on deep neural networks, which appear advantageous with large training datasets. The objective of this work is to provide a fully automatic images classification system.We propose an alternative representation of the input data called ultrasound images. Our approach is implemented on two different architectures of deep neural networks – Visual Geometry Group 16 (VGG16) and AlexNet for the classification of pulmonary pathologies. The ultrasound database was chosen as an input to classify three types of lung conditions – COVID19, Pneumonia and Non-COVID19. The proposed system has 53% for precision and accuracy using VGG-16 and 54% accuracy and precision using AlexNet, these results are taken with imbalanced data and without preprocessing for our data