Recently, researchers from Tel-Aviv University, University of Maryland School of Medicine, Baltimore, Mount Sinai Hospital, New York and RADLogics proposed AI-based automated CT image analysis tools for detection, quantification, and tracking of coronavirus disease. The motive behind this research is to develop deep learning-based automated CT image analysis tools and demonstrate that they can differentiate coronavirus patients from those who do not have the disease by detecting, measuring, and tracking disease progression.
How the System Works
The system proposed by the researchers receives thoracic CT images and flags cases suspected with COVID-19 features. For the cases which are classified as positive, the system outputs a lung abnormality localisation map and measurements.
The deep learning system is comprised of several components and analyses the CT case at two distinct levels, which are:
- 3D analysis of the case volume for nodules and focal opacities using existing, previously developed algorithms.
- Newly developed 2D analysis of each slice of the case to detect and localise larger-sized diffuse opacities including ground-glass infiltrates. According to the researchers, the ground-glass infiltrates have been clinically described as representative of the coronavirus.
The 3D analysis of the case volume uses commercial off-the-shelf software to provide quantitative measurements, including volumetric measurements, axial measurements (RECIST), HU values, calcification detection, etc. The next step, working in the 2D slice has a few advantages for deep learning-based algorithms in a limited data scenario. The advantages include an increase in training samples (with many slices per single case), using pre-trained networks that are common in the 2D space as well as easier annotation for segmentation purposes.
Behind the System
In order to detect the coronavirus related abnormalities, the researchers used a Resnet-50- 2D deep convolutional neural network architecture where the network is 50 layers deep and can classify images into 1000 categories. The network was pre-trained on more than a million images from the ImageNet database, and it was further fine-tuned by using suspected COVID-19 cases from several Chinese hospitals to solve the problem at hand. In order to overcome the limited amount of cases, the researchers employed data augmentation techniques such as image rotations, horizontal flips, and cropping to the network.
Advantages of This System
- AI-based automated CT image analysis tools can achieve high accuracy in the detection of coronavirus positive patients as well as quantification of disease burden.
- For coronavirus patients, the system outputs quantitative opacity measurements and a visualisation of the larger opacities in a slice-based “heat map” or a 3D volume display.
- This system will allow a greater volume of patients being screened for coronavirus in a shorter period of time.
Researchers from around the world have been trying to utilise the capabilities of AI to help accurately detect and track the progression. With the help of the deep-learning image analysis system developed, the researchers achieved classification results for coronavirus vs Non-coronavirus cases per thoracic CT studies of 0.996 AUC (95%CI: 0.989-1.00) on datasets of Chinese control and infected patients.
Read the paper here.
If you loved this story, do join our Telegram Community.
Also, you can write for us and be one of the 500+ experts who have contributed stories at AIM. Share your nominations here.
A Technical Journalist who loves writing about Machine Learning and Artificial Intelligence. A lover of music, writing and learning something out of the box. Contact: firstname.lastname@example.org