An article published in medical journal The Lancet on February 25 finds that reduced medical resource levels will trigger a spike in the coronavirus death rate in the local population beyond the current estimates. The study shows that death rates are over 3 percent in Wuhan city, 2.9 percent in Hubei province, while only 0.7 percent across the rest of China.
Close to 30,000 medical staff from across China have been dispatched to Hubei province to help overworked local medical professionals in the fight against COVID-19. Fast and accurate diagnosis is critical on the front line, and now an AI-powered diagnostic assessment system is helping Hubei medical teams do just that.
Currently, CT lung scans and nucleic acid tests are the two main diagnostic tools doctors use in confirming COVID-19 infections. CT imaging is crucial for diagnosis verification, and also allows doctors to access lung infection severity.
On Chinese New Year’s Eve (January 24) the number of confirmed cases in China passed 1,000 for the first time. As medical resources for infection screening became increasingly scarce, the Shanghai Public Health Clinical Center (SPHCC) reached out to AI startup YITU Technology’s healthcare division, which provides AI-powered medical imaging solutions for lung cancer diagnosis.
YITU’s development team cancelled their New Year’s plans and set out to build a AI CT image reader. “The R&D cycle of computer vision products from modelling to clinical application normally takes six months to one year, and requires follow up optimization on-site in hospitals,” explains YITU healthcare team general manager Su Xiaoming. The 100-strong YITU team was determined to dramatically shorten that process.
Working closely with doctors from SPHCC and under the risk of infection, YITU developers completed the first iteration of their AI diagnosis platform in just four days. They discovered that in comparison with ordinary pulmonary nodules shown in CT scans, more fluid lesions were found in coronavirus-infected patients’ CT scans, which show the following patterns and stages:
- Early lesions often show small patchy shadows and interstitial changes in lungs;
- As the lesions grow, multiple ground glass opacities and infiltrates appear in both lungs;
- In severe cases, patients have diffuse lung lesions and pulmonary consolidation may occur; and lungs will appear “white”.
Su told Synced that from January 28 to February 13 the YITU AI diagnosis model was applied in 659 cases at SPHCC, where the system’s coronavirus detection sensitivity reached 97.3 percent and its diagnosis specificity reached 99 percent.
By February 5th, the AI diagnostic system was deployed in four Hubei province hospitals struggling with ongoing doctor and supply shortages, including Renmin Hospital of Wuhan University.
In the hospitals, the diagnosis and treatment of coronavirus involves doctors from different clinical departments — radiology, respiratory, emergency, intensive care, etc. — and so the YITU medical team have built system functionalities catering to their different needs. Respiratory departments and ICUs for example need quantitative assessment of lesions, thus a lung histogram curve analysis was added to compare with normal human levels.
At the moment, the YITU “Coronavirus Chest CT Smart Evaluation System” can compress the diagnosis of suspected cases to 2-3 seconds. The company also designed a chatbot doctor that can help the public with self-diagnosis through Q&A and recommend nearby hospitals and appointment booking services.
Source: Synced China
Localization: Meghan Han | Editor: Michael Sarazen