NVIDIA Inception partner TrainingData.io recently developed a segmentation AI model that enables instant analysis of the progression of COVID-19 in chest CT images.
When the virus progresses through a patient’s body, there is a build-up of fluid in the tiny air sacs in the lungs called alveoli. The presence of this fluid causes inflammation of the lungs. The growth in inflammation of the lungs can be observed in X-ray and CT imaging. The inflammation of the lungs is visible in the form of ground-glass opacities (GGOs) that are followed by ground glass consolidations.
The Segmentation Model for COVID-19 Detection
TrainingData.io’s segmentation model for chest CT is built using NVIDIA Clara Train SDK. The segmentation model has two classes; a: lung, and b: COVID-19 infection. It can be accessed from their web-application without any prior setup. Once the results of the segmentation are generated, TrainingData.io enables further annotation at pixel level and visualization in 3D, allowing researchers and data-scientists to train their own segmentation model for detection of COVID-19 in chest CT.
Any researcher or data-scientist can go to TrainingData.io’s website, upload a chest CT study, and instantly visualize COVID-19 infection in chest CT as shown below.
Edge Case Discovery & Continuously learning AI using GPU Servers
TrainingData.io has an active-learning-data-pipeline that enables machine learning engineers to continuously improve ML models with a single click of a button. As the segmentation model analyzes new datasets the data-scientist can discover the edge cases, where the segmentation model failed, with the help of a trained radiologist. This process of discovering edge cases has been very cumbersome for the data-scientists. TrainingData.io’s workflow has streamlined the edge case discovery for the data-scientists.
TrainingData.io’s COVID-19 and lung segmentation models use NVIDIA Clara Train SDK for chest CT. With Clara, researchers and developers can take ML models and write an application workflow around it to enable interfacing in a hospital-like environment. Clara Deploy also provides researchers with a reference deployment pipeline which can be seamlessly evaluated for localized data. As newly annotated chest CT exams become available, the model can be customized and re-trained.
“NVIDIA Clara’s powerful features like Bring-Your-Own-Model (BYOM) and web-interface for inference, have enabled us to build an active-learning-data-pipeline for radiology,” said TrainingData.io Co-Founder and CEO Gaurav Gupta.
To power the annotation and training of their robust AI models, Trainingdata.io runs on NVIDIA T4 GPUs on a cloud service provider. Their highly detailed 3D visualizations also run on NVIDIA GPUs.