AI-Med: Artificial Intelligence at the service of human health
During the 5th EISA Summit, senior research and technology advisor of AI-Med research consortium presented his team’s “AI-based disease detection system,” a module of which is AI-based COVID-19 detection system.
MSTF Media reports:
Hamid R. Rabiee, professor at the Department of Computer Engineering, Sharif University of Technology, and a member of AI in Digital health at the WHO, in a pitch-deck during the 5th Exposure of Industries to Scientists’ Achievements Summit (EISA), presented the “AI-based disease detection system” he has developed along with his team. This is “a rapid platform of disease detection system with Explainable Artificial Intelligence (AI).”
He said that their team is comprised of “around 300 experts in different fields.”
“Having extensive experience in working on enterprise platforms, we have been working on AI-based platforms, not only solutions and services, for the past 15 years,” he added.
Rabiee stated that they have “9 different labs” in different fields of science and technology, including health.
In the field of health, they have been joined with “more than 5 medical schools” nationally, in order to design practical systems suitable for being used in different medical centers.
Rabiee remarked that the use of AI is growing in a very fast pace worldwide in different sectors of health—from designing medicines to diagnosis of different diseases.
“One immediate application of AI in health is using it as an assistant to radiologists for diagnosis of different diseases,” he added.
He emphasized that what he presents in this pitch-deck is “basically one service of our health platform” which is designed to diagnose different diseases based on medical data including images, clinical data, and other pieces of information.
“This system is designed to operate on a multimodal mode to detect different diseases on a timeline,” he continued.
He stated that because of the COVID-19 pandemic, they have focused on detection of COVID-19 infection using CT scan images.
Pointing out the necessity of adopting AI in the sphere of radiology, he stressed that the amount of imaging is skyrocketing, while the number of radiologists is struggling to catch up.
As Rabiee put it, “according to the WHO, there is a serious lack of radiologists across the globe. There are many countries that do not have sufficient number of radiologists.”
His multidisciplinary team has designed an AI-based COVID-19 detection system that recognizes anomalies that are visually hard to detect in chest CT scans.
“We started this project back in February 2020, when we decided to have a module on COVID-19 on our AI-Med platform,” he said.
“We collected around 2000 pictures in 4 weeks from different medical centers,” he continued.
Pointing to the first problem that they encountered in this project, he said “the images came from different devices with different parameters and different batch effects.”
In order to obtain accurate results, “we had to take care of the batch effects on the registration of the images as well,” he noted.
The other issue was “the fact that not every infection on the lungs is COVID-19. In other words, it might be a different disease, but the lungs are infected anyway,” he observed.
“So, we should have been able to diagnose whether the infection or anomaly is COVID-19 or another disease,” he continued.
Another problem they faced was “the time of processing,” he said, adding that they wanted to have the processing done “in less than one or two minutes.”
According to Rabiee, even though RT-PCR test kits are used widely around the world to detect COVID-19 infection, they are time consuming, prone to common laboratory errors, and have low accuracy.
His team has designed a system that detects the infection very fast; it does the diagnosis processing “in less than a minute,” he maintained.
“This system is available at a very low cost,” Rabiee announced, adding that many medical centers are provided with this system free of charge right now.
Moreover, this system is widely accessible; “it can be used remotely or can be easily integrated to the HIS and PACS systems in hospitals,” he noted.
Drawing upon deep learning, the team has been successful in designing a system that “not only can detect the diseased and healthy lungs, but also is able to recognize whether the lungs are infected with COVID-19 or other infections.”
According to Rabiee, it is of high important for the physicians to know “the volume of infected regions in the lungs” and that “which parts of the lungs are infected,” because they can administer effective treatments for patients based on the regions that are heavily involved.
“Our system provides the physicians with these specific data,” he continued.
So far, “more than 8000 cases” from different medical centers and hospitals have been tested by this system.
He said that this system can detect the infection in the early stages of disease, when it is not visible by human experts.
Although it was not initially attainable, now “we can predict whether a patient survives or not—with 88% to 90% accuracy,” he announced.
This option helps the medical centers “manage their resources much better” and operate much more efficiently, Rabiee observed.
He further hinted at other unique features of AI-Med, stating that “the overall accuracy score of our model is above 95% and the sensitivity is above 98%.”
“By no means are we saying that this system replaces radiologists; rather, it is a good assistant to the experts,” he emphasized.
Talking about AI-Med’s worldwide outreach, he said they have started some cooperation within the last few months with “Canada, India, and some countries in Europe like Romania.”
“We will also discuss the possibility of joint venture with Indonesia in two weeks,” he added.
Rabiee reiterated that the sphere of their project is not limited to the detection of COVID-19 only; they have a multimodal platform for detection of other diseases using different medical data as well.
“We are going through some procedures to get some international standards for this system,” he continued.
He stated that they are focused on getting the CE Mark for the platform, while getting ISO 13485 certificate is in progress. They are also planning to get the FDA’s approval for the platform as an AI platform for diagnosis.
“We have started to add breast cancer module. We are currently doing the research, but it is expected to be available in March 2021,” he announced, adding that they have plans to add “lung nodules detector” to the system later.
AI-Med is a multidisciplinary research consortium utilizing Artificial Intelligence technologies to improve the quality and accuracy of digital health services.
EISA, a platform composed of various commercialization services in the STI Eco-system, is developed by the Mustafa Science and Technology Foundation to meet the needs of the Islamic world’s knowledge-based industry and scientific community, enabling technological achievements in products and services to be introduced to new markets, industries, and investors. The 5th round of EISA focuses on “Cutting-edge Healthcare and Medical Achievements.” This round is held virtually from November 30th to December 7th, 2020.
