Building a clearer picture

How are covid-19 patients diagnosed? It’s a process that involves several aspects: epidemiological history, clinical manifestations, hematological changes and chest imaging.

Epidemiological history determines whether you have traveled to virus-affected locations, or have been in contact with someone infected.

If a case is suspicious, then a reverse transcriptase-polymerase chain reaction test is performed. If the antigen test detects the presence of the Sars-Cov-2 in one’s body, they are diagnosed with Covid-19.

But there are many problems with RT-PCR testing, such as false negatives and a time lag in diagnosis, said Yin Guosheng, the head of the department of statistics and actuarial science at the University of Hong Kong.

“The test, which takes a swab from the nose or throat for a trace of the virus, sometimes requires several trials to make a final confirmation,” he said. “That would put patients at a great disadvantage, as they cannot be diagnosed quickly and be provided with the necessary quarantine and treatment at an early stage.”

This is why chest CT scans play another vital role in the diagnosis of Covid-19, especially since it is useful among those with none or minimal symptoms, as the coronavirus will typically first attack the lungs, causing lesions.

But there are also downsides to CT scans – they can be inadequate in observing relatively hidden lesions during the early stages and distinguishing between Covid-19 and other forms of pneumonia.

Chest CT scans are also time-consuming, requiring about four to five hundred images for each adult.

To address these challenges, a research team led by Yin and Liu Bin, an HKU post-doctoral fellow and assistant professor at the school of statistics at Southwestern University of Finance and Economics, has come up with a solution – using integrated radiography and computer vision to develop an online diagnostic system for Covid-19 based on chest CT scans.

The system can help screen suspected cases of Covid-19 and quickly evaluate the probability of infection. The diagnostic result is immediate and has an accuracy of 90 percent.

But the system’s advantages go beyond that.

Other than diagnostic issues, scientists also struggled with analyses of CT scans.

“Since most of the chest CT datasets of Covid-19 patients are not publicly shared, we have to spend much time to search for publicly available samples and tag them,” Yin said.

That is why the diagnostic system by Yin and his team was designed as an open-sourced user-friendly website with all codes and data freely available.

Early studies in the role of CT scans as a method of Covid-19 diagnosis showed that scan results were unable to meet clinical standards.

Apart from a small sample size, the reason behind this is believed to be that the rich annotations associated with the CT images have not been fully utilized.

“Most of the research papers do not share data and computer codes, and this does not facilitate knowledge exchange and disease prevention around the globe,” the two researchers said.

In addition to providing an open-sourced database, the data system also offers a solution for distinguishing Covid-19 from other similar diseases. Unlike current CT images and traditional medical imaging datasets, each of the new system’s CT samples is collected from a research preprint.

In it, clinical experts have comprehensively annotated the CT images of Covid-19 patients with detailed descriptions of lesions.

The team has analyzed and pinpointed five distinctive lesions associated with Covid-19, which are the unique features that differentiate the virus from other lung diseases. Studying 760 papers, it has been established that each Covid-19 patient will develop at least one of these five lesions.

As clinicians tend to focus on abnormal lesion areas when analyzing CT images, the lesion-attention deep neural network model focuses on the primary task of binary classification for Covid-19 diagnosis, while an auxiliary multilabel learning task is implemented to draw the model’s attention to the five lesions.

Yin explained that as the auxiliary task assists the primary task to focus on the lesion areas, the diagnostic accuracy can be improved drastically.

In hopes that medical staff can benefit from the diagnostic system and share patient image data to initiate collaborative research and accommodate the urgent demands for Covid-19 testing, the team said it will continue to collect new samples and improve the training model periodically.

(This article was published at The Standard on June 9, 2020: Education: Building a clearer picture )

200609 Building a clearer picture

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s