COVID CT Scan Classification: ResNet50 vs. EfficientNet-B0

Despite the waning presence of COVID, CT scan images captured during the pandemic persist as a crucial resource for deep learning practitioners in medical imaging. In my recent Image Classification project, I delved into the realm of multi-class classification using two influential deep learning models, ResNet50 and EfficientNet-B0. In this post, I present a detailed analysis of my recent Image Classification project, where I utilized two popular deep learning models, ResNet50 and EfficientNet-B0, to classify COVID CT scans.

Dataset Overview

The dataset consists of a diverse range of CT scans, including:

Healthy scans: 758 SARS-CoV-2 infected scans: 2,168 Other pulmonary conditions: 1,247

Models Used

1. ResNet50

• Training Accuracy: 94.48%
• Testing Accuracy: 87.81%

2. EfficientNet-B0

• Training Accuracy: 98.95%
• Testing Accuracy: 92.14%

Results and Insights

ResNet50 Performance

ResNet50 model achieved an impressive training accuracy of 94.48% and a testing accuracy of 87.81%. This indicates robust learning capabilities and generalization to unseen data. The model excelled in distinguishing between healthy scans, SARS-CoV-2 infected scans, and other pulmonary conditions.

EfficientNet-B0 Performance

The EfficientNet-B0 model outperformed ResNet50, achieving a training accuracy of 98.95% and a testing accuracy of 92.14%. This suggests that the model has a higher capacity to learn intricate patterns and features within the dataset. The enhanced performance could be attributed to the model’s architecture, which optimizes both depth and width.

Code Contribution

To foster collaboration and transparency, I have made our Python code publicly available on two prominent platforms:

GitHub: Here

Kaggle: Here