AI that's built to save lives

CBRC team led by Associate Director Xin Gao is putting AI to use to help the Kingdom's radiologists more easily and rapidly identify COVID-19 cases.

​Associate Professor Xin Gao and his group have developed an artificial-intelligence (AI) based solution to help increase COVID-19 testing accuracy. Identifying cases of early stage infection has been particularly challenging for frontline clinicians. Gao's AI-based model, which aims to increase accuracy, has been put to immediate use at King Faisal Specialist Hospital (KFSH) in Riyadh.

"The model was fast to use and each case took approximately less than a minute to be processed. It is expected that such a model will make an important contribution to chest imaging, especially with the current pandemic," said Dr. Riham Eiada of the King Faisal Specialist Hospital.

Supporting clinicians with greater accuracy

To date, the gold standard for confirmation of COVID-19 has been nucleic acid detection. Unfortunately, this method of testing alone has had a high, false-negative rate, especially for patients in the early stages of the disease.

In response, Gao and his group in the Computational Bioscience Research Center (CBRC) proposed a fully-automatic, rapid, accurate, and machine-agnostic method that can segment and quantify the infection regions on CT scans from different sources.

Despite urgent advances in the development of AI-based computer-aided systems for CT-based COVID-19 diagnosis, whereas the existing segmentation methods require a high level of human intervention. The KAUST team has resolved these shortcomings with three novel innovations.

Three novel innovations

The first is a novel embedding strategy. Gao's team has developed a way to project any CT scan into the same, standard space, which makes the model machine-agnostic.

The second innovation has been the development of the first CT scan simulator for COVID-19. The team found that fitting the dynamic change of real, patient data, measured at different time points, resolves the data scarcity issues that plague traditional approaches.

Third, the team has created a novel 2.5D, deep-learning algorithm to solve the large-scene-small-object problem. By decomposition of the 3D segmentation into three 2D ones, they are able to reduce the overall complexity of the model by an order of magnitude and, at the same time, dramatically improve segmentation accuracy.

Xin and his team conducted comprehensive evaluation and validation on multi-country, multi-hospital, and multi-machine datasets, which demonstrated the superior performance of their method over existing ones and suggests it is important application value in combating the disease. The system is currently being deployed to a number of hospitals for prospective pilot studies.

"The system has been tested on 32 scans," a KFSH radiologist said. "Only two cases were not accurately detected areas by AI. The model was 100% accurate in the remaining cases."

Related stories: