Original article was published by Sritan Motati on Artificial Intelligence on Medium
How It Works
A group of researchers has developed an AI algorithm that was trained to classify COVID-19 pneumonia in computed tomography (CT) scans of people’s lungs. Let’s take a closer look at the two main components of this algorithm: CT scans and the deep learning-based program.
CT scans reveal clear features of lungs with COVID-19 pneumonia, which makes the AI’s job of classifying the scans as infected or not infected much easier. Additionally, they can detect COVID-19 in people without symptoms, with early symptoms, while symptoms have peaked, and after symptoms have been resolved. The often-used alternative to these scans, reverse transcription-polymerase chain reaction (RT-PCR) tests, pose many challenges, such as classifying positive cases as negative at a fairly high rate, which makes CT scans a better option sometimes.
Unfortunately, CT scans do not distinguish COVID-19 pneumonia from influenza-associated pneumonia, which is why they’re not always recommended to diagnose pneumonia. To combat this, the AI algorithm was programmed to make sure it only diagnoses COVID-19 pneumonia. How? With deep learning.
Deep learning is a subset of artificial intelligence that uses algorithmic structures called neural networks to perform classification tasks. Deep learning models learn to classify things from lots of input data, whether it be images or text, so to train this model, the researchers collected 2,734 CT scans from hospitals around the world.
The model had two main parts: a lung segmentation algorithm that identified lung regions in the CT scan and several image classification models that used the lung segmentations to classify the lungs as infected or not infected.