EECS545_FinalProject
This project is about "COVID-19 X-Ray Image Classification Using Transfer Learning and Contrastive Learning". As COVID-19 spreads around the world, increasing demand for COVID-19 testing burdens health care workers. Researchers started to utilize machine learning techniques to identify COVID-19 cases using X-rays for fast and more accurate predictions to minimize the spread. Motivated by this, a decision fusion method using multiple CNN models has been proposed in this research to accelerate the analysis of chest X-ray images. We have leveraged the power of three pre-trained models (VGG16, InceptionV3, DenseNet121) with freezing of the feature extraction layers to extract the features of the image datasets and trained the last fully connected layers to classify the images. Moreover, We present a Siamese network to integrate the Contrastive learning with a fine-tuned pre-trained DenseNet121 model to capture unbiased feature vectors for final classification. Finally, We have achieved 95.57% accuracy in segmented lung test images using the idea of Majority voting. We have also used Grad-Cam heatmap visualizations to prove the necessity of applying "lung segmentation" to the dataset, to have a robust classification model.