RT-PCR testing resources are limited, turnaround times can be lengthy, and reported sensitivities vary. In response, UZ Brussel is making use of low-dose CT imaging, reported to have high sensitivity for the detection of COVID-19, but low specificity (97% and 25% with RT-PCR as reference, respectively). However, in the current epidemic outbreak of the COVID-19, CT is considered an accurate detection tool (Tao 2020). CT is currently employed for several aims, including (i) triage of symptomatic patients for referral to RT-PCR
testing, which relies on evaluating the percentage of affected lung tissue and number of lobes. Based on the associated score, a decision is taken on hospital admission, referral to RT-PCR testing and likely-hood of COVID19

infection (important in deciding on which ward the patient is received). This quantification of affected
tissue is cumbersome and associated with low inter-observer agreement.
An automated approach for this quantification step may save valuable time. In particular, the reproducibility
of the approach is expected to also enable more detailed assessment of a second aim, namely (ii) the
monitoring of disease progression. Related AI-approaches, tuned for sensitivity, are expected to aid in (iii)
the detection of small initial lesions for asymptomatic patients.
In term, given sufficient data and effort, we believe CT combined with AI can play a pivotal role in
distinguishing COVID-19 from other pathologies, the first evidence of which has recently been reported (Li
2020). The latter would enable (iv) CT as a diagnostic tool for COD-19. Finally, such large-scale analysis may
enhance insight, provided (v) explainable AI techniques are employed enabling human understanding of
predictive patterns.
Effective start/end date1/05/2031/10/20

    Flemish discipline codes

  • Medical imaging and therapy not elsewhere classified

    Research areas

  • Covid 19

ID: 51828747