The COVID-19 Pandemic has had a disruptive effect on the world. Since its emergence in late 2019, the virus has brought the global economy to a standstill.
More importantly, the virus has claimed the lives of hundreds of thousands around the world, with the number of infected rising into the millions. In the People’s Republic of China alone, the virus has infected over 80,000 people and killed more than 4,500.
Healthcare systems in mature and emerging markets are buckling under the pressure of this global crisis. Though many of those who have been infected have received treatment, healthcare professionals, tools, and resources are unprecedently strained. And many more infected people may potentially be going undetected and untreated.
Early detection is key to mitigating the pandemic
In this context, early and accurate detection is an absolute priority for governments the world over.
Antibody tests are becoming more widely available, but for now their speed and convenience is trumped by their low accuracy. Meanwhile, polymerize chain reaction (PCR) tests conducted in laboratories are labor intensive, with several stages at which errors – and false negatives – may occur.
As we learn more about the virus and the way it impacts the human body, more can be done to ensure that fewer people fall through the cracks.
AI-assisted imaging for better diagnoses
Huawei CLOUD’s Medical Imaging AI is collaborating with its partners Lanwon Technology and HY Medical to address this issue.
It leverages the EIHealth platform – a professional R&D platform designed to accelerate research in new applications for AI in genomics, drug discovery, and medical imaging – to develop and deploy an AI program specifically designed to diagnose COVID-19 in CT scans of patients’ lungs.
CT scans are fast and accurate but need to be rechecked and reviewed multiple times over short periods of time because of the large number of lesions caused by COVID-19, and the often-rapid changes the virus creates within the lungs. This can significantly increase the workloads of imaging specialists who are often short in supply at most hospitals.
The AI-assisted Patient Screening system for COVID-19 relies first on the guidance of doctors who label the tell-tale lesions caused by COVID-19-induced pneumonia in CT scans of patients’ lungs. Labelled CT scans are then used by AI scientists at Huawei CLOUD to train an algorithm to independently identify similar symptoms in other CT scans.
The algorithm uses computer vision and medical imaging analysis to quickly and accurately output CT quantification results. Having analyzed hundreds of infected and non-infected patients, the algorithm uses Dice Similarity Coefficient (DICE) statistical analyzes and Absolute Volume Difference (AVD) imaging technology.
DICE and AVD are used in concert to automatically segment lesion areas: DICE overlaps predicted and actual lesions, while AVD identifies the differences in volume between them. The results are consistent with doctors’ manual sketching.
The screening system is then able to automatically identify and digitally segment lung lesions and measure lesion volume. This allows the rapid diagnosis of symptomatic patients carrying the COVID-19 virus – up to three days faster than other testing methodologies.
Supporting and complementing doctors’ work
According to Huawei CLOUD Healthcare Manager Mr Zhang Shaowei, “The components of the EIHealth platform use different AI techniques, including Machine Learning, Deep Learning, and Knowledge Graph Learning to power a screening system which can autonomously and holistically integrate the lessons learned by our medical professionals.”
This process also boasts an impressively high accuracy rate, correctly identifying carriers 98% of the time. Using this technique, doctors can delegate the task of reading CT scans to the AI and focus instead on the important work of treating patients.
Furthermore, for confirmed cases in hospitals, the AI-Assisted Patient Screening system can quickly perform registration and quantitative analysis on the data acquired over multiple rechecks of the CT scans, thereby helping doctors evaluate patients’ conditions and the impact of targeted drug treatments.
As Mr Zhang notes, “The system does not fully replace the work that doctors do, but enhances their capabilities by adding an additional line of defense in the diagnostic process.”
Exporting the expertise
Huawei CLOUD has already deployed the AI-Assisted Patient Screening system in 60 hospitals across the globe, including Malaysia, Thailand, Italy, and Ecuador.
In Malaysia, Huawei is working with the Ministry of Health to deploy the system in the Sungai Buloh Hospital (state of Selangor). Dr Yun Sii Ing, Head of Department for Clinical Radiology at the hospital, recently stated that, “[…] This equipment (…) will help save lives by greatly reducing the risk of infection and enabling medical personnel to perform their duties better and faster.”
In Ecuador, the General Hospital of South Quito was connected to Huawei’s AI system, enabling diagnoses just 14 hours after its launch. Otto Sonnenholzner, Vice President of Ecuador, took to social media to thank Huawei for making Ecuador the first Latin America territory with AI-powered diagnostics.
Beyond the COVID-19 pandemic
Huawei CLOUD’s algorithm holds promises beyond the current crisis.
Designed to adapt quickly to the type of data it receives, the AI program can be fed new information on changing or evolving symptoms, allowing the system to tailor its delivery of diagnoses accordingly. This quality is vital, as new information about the virus is continuously coming from on-the-ground reports, and any delay in processing this information can cost human lives.
The system will also contribute to more long-term research on the virus. The data that is generated by the AI-Assisted Patient Screening system is usable in quantitative analyzes and can be used to generate 3D reconstructions for further assessment. Data can also be easily exported and utilized for further quantitative tasks, as well as shared between medical teams.
The speed at which Huawei CLOUD’s existing platform was configured to address a new and emerging problem is also a testament to the progress AI scientists have made in introducing flexibility into the design process of AI platforms.
Mr Zhang sees this flexibility as a key driver of AI’s ability to address broader humanitarian challenges in the near future.
“AI scientists are getting better at adapting algorithms and using AI to solve new and more difficult problems,” he says. “The work of Huawei and other AI companies during this pandemic is proving that AI-powered solutions will play an increasingly larger role in a wide range of disaster-management strategies in the future.”