How Machine and Deep Learning Models Help For Covid-19 Diagnosis?

Introduction

With the ever-growing demand of screening millions of possible COVID-19 cases and because of the advent of high false negatives in typically used PCR tests, the necessity for searching for an alternate simple screening mechanism for COVID-19 Diagnosis using radiological images, such as chest x-rays, has gained importance. In such a scenario, Machine Learning (ML) and Deep Learning (DL) provide quick, effective, and automated approaches to identify the irregularities and derive key features of the altered lung parenchyma, which may be linked to specific signs of the COVID-19 virus.

The need of the hour is early and quick detection, and this is all the more important as the healthcare system is overwhelmed with the flood of patient data as the days are progressing. 

How can Machine Learning and Deep Learning Help?

Chest x-rays look similar for COVID-19 patients, with lungs appearing hazy and patchy rather than healthy and clear. However, heart failure, pneumonia, and other chronic illnesses in the lungs can also look similar to COVID-19 in x-rays. Trained radiologists should be able to tell the difference between a less contagious disease and COVID-19.

To understand if ML and DL can help with diagnosing COVID-19 patients, researchers at the Northwestern Memorial Healthcare System used 17,002 chest x-rays to create, train, and test a ML based algorithm. Of those images, 5,445 came from COVID-19 positive patients. 

The team then tested the ML algorithm against five experienced cardiothoracic fellowship-trained radiologists on 300 random test images. Each radiologist took approximately two-and-a-half to three-and-a-half minutes to examine this set of images, while the ML model took about 18 minutes.

The results showed that the ML model performed slightly better with an accuracy of 82 percent, while the radiologists had an accuracy of 76 – 81 percent. The findings indicate the ability of the ML model to provide quick and more accurate diagnoses than those that are typically delivered with standard care.

Thus, it is safe to say that ML tools could be a better alternative in hospitals and health systems during the resource-intensive pandemic.

Radiologists are not always available and are also expensive,” states Aggelos Katsaggelos, an AI expert and senior author of the study. “X-rays are reasonable and already a typical element of routine care. This could possibly save money and time – especially because timing is so vital while trying to manage the pandemic.

Can Deep Learning and Machine Learning Completely Replace Manual Testing?

However, the DL and ML platforms are not meant to substitute the standard methods of diagnosis entirely. Because not all COVID-19 patients show indications of an illness – even in chest x-rays – the system does not diagnose those patients with the virus. 

ML tools could still help physicians quickly screen patients who are admitted at hospitals for reasons other than COVID-19. The ML model could provide fast and quick detection of the virus, which could in turn, prevent patients and providers by detecting the patients’ need to isolate sooner. “In such cases, the AI system will not flag the patient as positive,” Wehbe states. “But neither would a radiologist. Clearly, there is a limit to a radiologic diagnosis of the COVID-19, which is why it cannot be used to replace testing entirely.”

The researchers have launched the algorithm to be available publicly, so that others can keep training it with new data. The model is still in the research phase, but could potentially be used in clinical settings in the future.