IHS Markit (now a part of Informa Tech) speaks with Dr. Matt Deng, Chief Scientist and Director of Infervision North America, about the company’s targeted AI which has already been used in 10,000 coronavirus cases.
As of Feb 18, there are 73,332 confirmed cases of the coronavirus disease (COVID-19) in 25 countries, 1,901 of these cases are new (World Health Organization). The number of confirmed cases jumped yesterday when the WHO began including both laboratory-confirmed and clinically diagnosed cases. Previously, the WHO was not including over 13,000 clinically diagnosed cases that were recently reported from Hubei. For the first time during this epidemic, China had reported clinically diagnosed cases in addition to laboratory-confirmed cases. To maintain consistency in tracking the outbreak, the WHO was only reporting on the number of laboratory-confirmed cases while waiting for more information such as when the cases occurred during the outbreak and whether suspect cases were reclassified as clinically diagnosed cases. As the scale of the outbreak continues to grow, the ability to effectively triage is even more essential in curtailing the spread of the disease.
One of the WHO’s strategic objectives is to prevent further international spread from China, which can be achieved through a combination of public health measures, such as rapid identification, diagnosis and management of the cases. Rapid identification can be enabled with AI tools at the clinical level, rather than waiting for lab results. Infervision’s Coronavirus AI solution has been tailored for this front-line use and is accelerating pneumonia diagnosis and epidemic monitoring efforts at the Zhongnan Hospital of Wuhan University, Tongji Hospital in Wuhan, and at sites in other cities such as the Third People’s Hospital of Shenzhen.
Diagnosing the Coronavirus
The typical diagnosis protocol may include a physician ordering laboratory tests (RT-PCR) on respiratory specimens and serum to detect the disease, or a CT scan may be used to identify chest infection. However, the surging number of patients is creating thousands of studies to read daily and the results are taking hours to be delivered. Further, the limited supply and time-consuming nature of DNA tests, along with their susceptibility to false negatives, underscore the need for improved screening for the early detection of COVID-19—a barrier that CT and AI may resolve. Infervision’s Coronavirus AI is helping hospitals, particular those with limited resources, to quickly screen suspected patients for further diagnosis and treatment. In as little as two minutes from the time a CT scan is conducted, the radiologist is notified whether the patient needs to enter the care protocol for the coronavirus.
According to Dr. Matt Deng, Chief Scientist and Director of Infervision North America, Infervision’s Coronavirus AI is highly accurate with 95% sensitivity. The RT-PCR panel only has 30-50% sensitivity and requires a few hours to process, so is not necessarily the sole best tool for initial diagnosis. Infervision developed the Coronavirus AI in two weeks using real cases for training, and building on another algorithm that had been in development for pneumonia detection. To date, Infervision’s new AI tool has been used in 10,000 cases in China. Use in other countries is not in discussion yet, but may become necessary if an actual outbreak occurs beyond China.
More about Infervision
Based in Beijing, Infervision was founded in 2015, has raised $70 million in startup financing and is working with more than 350 hospitals, 50 of which are outside of China. Nearly all of China’s top-tier hospitals use its lung-specific AI tool. The AI portfolio from Infervision includes InferRead Stroke CT.AI which locates and quantifies area of intracranial hemorrhage on head CT images, InferRead Lung CT.AI which locates suspicious lung nodule cancer features from chest CT images, and InferRead CT Chest which enables the review of an image only once to perform multiple disease screenings in the chest. These screenings include lung nodule screening, chest fractures, bone metastases and bone tumor screenings, chronic lung disease (such as emphysema) screening, and cardiac calcification screening. Infervision’s machines are pending FDA approval, which was expected by 2020. However, regulatory approval may be delayed due to the coronavirus.
The market for diagnostic AI
IHS Markit (now a part of Informa Tech) has researched more than 290 vendors and 365 machines to provide a taxonomy of the ecosystem supporting machine learning with diagnostic applications. The vendors include companies, academic organizations, and healthcare providers. They are focused on aggregating and analyzing large datasets of information from clinical care services for the purpose of improving clinical decision support and health outcomes, not business analytics. Of the machines researched, 28% have regulatory approval (FDA, CE, MFDS, CFDA). This is up from 15% in 2018, but there are still many vendors currently in the process of obtaining approval, or they believe that their implementation of machine learning falls outside of regulatory oversight (at least 12% of these machines do not require approval). Venture funding for companies involved in machine development comes to more than $8.1 billion, including the latest information from all publicly available funding rounds. This does not include internal company investments, and excludes companies that have been acquired or completed an IPO.
The following companies have developed machines specifically for the chest and lung, and have received regulatory approval; Aidoc, Arterys, GE Healthcare, Imbio, Lunit, Oxipit, Quibim, Qure.ai, Riverain Technologies, Siemens, Thirona, Vida Diagnostics, VoxelCloud, and Zebra Medical Vision. For a list of all diagnostic machines with regulatory approval, see the following infographic ‘Artificial Intelligence (AI) for medical diagnostics’.
Also, Bhvita Jani, senior analyst with IHS Markit, recently discussed how AI powered recognition in CT screening will play a fundamental role in driving case prioritization and identification of key indicators and symptoms of the coronavirus in her article ‘How can the latest breakthroughs in medical imaging assist in diagnosis of the coronavirus epidemic? In the article Ms. Jani notes that “In such circumstances, AI powered recognition to drive case prioritization and identification of key indicators and symptoms of the coronavirus, in particular pneumonia, will provide solace in helping medical professionals to tackle and control outbreak of this global epidemic.”