Computed Tomography for COVID-19 Diagnosis

The outbreak of the novel coronavirus SARS-CoV-2, initially epicentred in China and
responsible for COVID-19 pneumonia has now spread to France, with 7730 confirmed cases and
175 deaths as on March 17th. Diagnosis relies on the identification of viral RNA by
reverse-transcription polymerase chain reaction (RT-PCR), but its positivity can be delayed.
A series based on 1014 chinese patients reported higher sensitivity for CT, with a mean
interval time between the initial negative to positive RT-PCR results of 5.1 ± 1.5 days
(PMID: 32101510). Moreover, obtaining RT-PCR results requires several hours, which is
problematic for patients triage.

Chest CT can allow early depiction of COVID-19, especially when performed more than 3 days
after symptoms onset. It is important to distinguish between COVID-19 and bacterial causes of
pulmonary infection, which requires expertise in thoracic imaging. Thus, it is important to
identify reliable CT diagnostic criteria based on visual assessment, and also develop
deep-learning based solutions for early positive diagnosis which could be used by less
experienced readers, in a context of large epidemic.

Several risk factors for poor outcome are already identified, such as older age,
comorbidities, or an elevated d-dimer level at presentation (PMID: 32171076). Extensive CT
abnormalities are linked to poor outcome, but some patients secondarily worsen despite non
extensive abnormalities at first assessment, highlighting the need for worsening prediction
based on initial imaging findings. Lastly, there is currently no drug with a proven efficacy
for patients with acute respiratory distress syndrome, who for management relies on
mechanical ventilation and supportive care. Some hypothesized that Remdesivir, an antiviral
therapy could be effective (PMID: 32147516), with ongoing randomized trials conducted in
China and the US. Automated tools allowing quantifying the disease extent on CT would be
desirable in order to evaluate the efficacy of new treatments.

Building a large dataset of CT images is needed for identification of accurate CT criteria
and development of deep learning-based solutions for diagnosis, quantification and prognostic
estimation.

The aim of this project is three fold: (i) create a multi-centric open database repository on
CT scans relative to COVID-19, (ii) create a multi-expert annotation protocol with different
level of annotations depicting the severity of the disease, (iii) allow the development of
non-proprietary computer aided solutions (academia & industry) for automatic quantification
of the diseases and prognosis through the use of the latest advances in the field of
artificial intelligence.

For patients, the validation of reliable diagnostic criteria will allow early detection of
the disease, and better distinction with other potential cause of acute respiratory symptoms,
requiring a specific treatment, such as bacterial bronchopneumonia. It will contribute to a
standardization of care as well as an equal access to diagnosis and treatment for the
ensemble of the population.

Public health benefit will be an access to CT diagnosis of COVID-19 independently from the
availability of local expertise in thoracic imaging. The possibility to anticipate the need
for ventilation, based on the developed CT severity scores, will also positively impact the
management of patients in particular in the context of a massive flow of patients as expected
at the epidemic peak. This project will allow evaluating the proportion of patients likely to
present respiratory sequelae, based on the severity and extent of lung abnormalities at the
acute phase of the disease.

The availability of automated quantification tools will help evaluating treatment efficacy if
new therapeutic approaches are developed.

Lastly, the developed tools for early diagnosis, evaluation of severity and prediction of
outcomes could prove useful if other viral pandemic occurs in the future. Indeed SARS-Cov2
outbreak has been preceded by SARS and MERS outbreaks due to other coronavirus.