Authors: Robert Turnbull, Simon Mutch

Published on: March 20, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2403.13509

Summary

• What is new: This paper presents a novel approach for detecting COVID-19 in CT scans through preprocessing and model ensemble techniques, and addresses domain adaptation for improved performance across different data distributions.
• Why this is important: The challenges include accurately detecting COVID-19 from CT scans and adapting the model to work with CT scans from different distributions.
• What the research proposes: The solution involves preprocessing CT scans to segment the lungs, training 3D ResNet and Swin Transformer models on these segmented images, and using an ensemble approach for annotation and fine-tuning.
• Results: The approach achieved a mean F1 score of 93.39% for detecting COVID-19 in Challenge 1 and 92.15% for Challenge 2’s domain adaptation task.

Technical Details

Technological frameworks used: nan

Models used: 3D ResNet, Swin Transformer

Data used: COV19-CT-DB database

Potential Impact

Healthcare providers, medical imaging software companies, AI diagnostic tool developers

Want to implement this idea in a business?

We have generated a startup concept here: RespiroAI.