CAD4COVID for rapid COVID-19 triaging
Early detection of COVID-19 is crucial to control the pandemic. Chest radiography (CXR) may play an important role in triaging for COVID-19, particularly in low-resource settings. Preferred tools in the detection of COVID-19 are RT-PCR testing and a CT scan, but in resource-constrained settings where the availability of such tools is limited and where COVID-19 is highly prevalent, X-ray can support rapid triaging. In many countries, X-ray is currently used as a first-line triage and COVID-19 disease severity indicator, before any further testing. CAD4COVID-XRay is a deep-learning AI system for the detection of COVID-19 characteristics on frontal chest radiographs benefiting from more than 10 years scientific CAD4TB R&D. According to a recent publication, the system outperformed all readers at their highest sensitivity for detection of COVID-19 characteristics.*
CAD4COVID a pro-bono initiative
CAD4COVID is a new artificial intelligence software to support the triaging of presumptive COVID-19 cases on digital chest radiographs by indicating and quantifying the affected lung tissue. This AI was developed as a joint research by Radboud University, various European hospitals, Thirona and Delft with the objective to facilitate rapid triaging for COVID-19 in resource-constrained and high-prevalence settings.
CAD4COVID is available since April 2020 and its use in the cloud is free-of-charge and can - depending on the diagnostic algorithm applied – be operated next to CAD4TB. In case of limited internet access offline CAD4COVID use can be provided through a compact Box. CAD4COVID is also developed for rapid CT image interpretation.
CAD4COVID-XRay intended use
The intended CAD4COVID's use – after symptom screening - is:
- to automatically detect and indicate in the heat-map the location of abnormalities in the chest radiograph suggestive of COVID-19
- to assess the disease severity through the abnormality CAD score (0 – 100) to allow care providers taking more informed decisions on e.g. testing, self-quarantine or hospitalization of patients in specialized ward for care.
- to quantify the percentage of visible affected lung tissue to help track disease severity and patient recovery
COVID-19 and TB combined AI use
The combined use of CAD4COVID and CAD4TB is crucial if programs want to respond to the growing concerns on significantly reduced TB case notification as reported by Stop TB Partnership recently:
With attention focused on coronavirus, undiagnosed and untreated TB cases will cause 1.4 million to die, research suggests
Up to 6.3 million more people are predicted to develop TB between now and 2025 and 1.4 million more people are expected to die as cases go undiagnosed and untreated during lockdown. This will set back global efforts to end TB by five to eight years.
All NTPs mentioned that they observe a decrease in the number of people presenting/accessing services for TB. In India, there is approximately an 80% decline in daily TB notifications during the lockdown period compared to the average daily notifications; the only country we had real-time data to analyze. Lack of real-time data makes it difficult even for national authorities to assess the situation in terms of decrease in case notifications and access. Other situations observed are:
- People avoid going to or have no possibilities to reach hospitals and medical clinics due to lockdown and reduced transportation means.
- Disruptions in sputum transportation and on providing different types of treatment support.
- Active TB Case Finding (ACF) activities have stopped.
- Disruption on diagnosis activities, due to lack of staff and sometimes even laboratory space.
The above indicates that COVID-19 may cause the rapid spread of TB in the near future as a reduced number of infectious cases is notified, hence not put on TB treatment so further spreading the disease. FIND and Swiss Tropical Institute perform operational research on diagnostic algorithms that include next to symptom screening AI for TB and COVID-19 using the same chest radiograph.