Skip to Main content Skip to Navigation
Reports

AI-based multi-modal integration (ScanCov scores) of clinical characteristics, lab tests and chest CTs improves COVID-19 outcome prediction of hospitalized patients

Nathalie Lassau 1, 2 Samy Ammari 1, 2 Emilie Chouzenoux 3 Hugo Gortais 4 Paul Herent 5 Matthieu Devilder 4 Samer Soliman 4 Olivier Meyrignac 4 Marie-Pauline Talabard 4 Jean-Philippe Lamarque 1, 2 Rémy Dubois 5 Nicolas Loiseau 5 Paul Trichelair 5 Etienne Bendjebbar 5 Gabriel Garcia 2 Corinne Balleyguier 1, 2 Mansouria Merad 6 Annabelle Stoclin 6, 7 Simon Jegou 5 Franck Griscelli 6 Nicolas Tetelboum 2 Yingping Li 1 Sagar Verma 3 Matthieu Terris 3 Tasnim Dardouri 3 Kavya Gupta 3 Ana Neacsu 3 Frank Chemouni 6 Meriem Sefta 5 Paul Jehanno 5 Imad Bousaid 6 Yannick Boursin 6 Emmanuel Planchet 6 Mikael Azoulay 6 Jocelyn Dachary 5 Fabien Brulport 5 Adrián González 5 Olivier Dehaene 5 Jean-Baptiste Schiratti 5 Kathryn Schutte 5 Jean-Christophe Pesquet 3 Hugues Talbot 3 Elodie Pronier 5 Gilles Wainrib 5 Thomas Clozel 5 Fabrice Barlesi 6 Marie-France Bellin 1, 4 Michael Blum 5
1 BIOMAPS - Unité BioMaps
2 Département d'imagerie médicale [Gustave Roussy]
3 OPIS - OPtimisation Imagerie et Santé
Inria Saclay - Ile de France, CVN - Centre de vision numérique
4 AP-HP Hôpital Bicêtre (Le Kremlin-Bicêtre)
5 Owkin France
6 IGR - Institut Gustave Roussy
7 DSA - Département de soins aigus [Gustave Roussy]
Abstract : The SARS-COV-2 pandemic has put pressure on Intensive Care Units, and made theidentification of early predictors of disease severity a priority. We collected clinical,biological, chest CT scan data, and radiology reports from 1,003 coronavirus-infectedpatients from two French hospitals. Among 58 variables measured at admission, 11clinical and 3 radiological variables were associated with severity. Next, using 506,341chest CT images, we trained and evaluated deep learning models to segment thescans and reproduce radiologists' annotations. We also built CT image-based deeplearning models that predicted severity better than models based on the radiologists'reports. Finally, we showed that adding CT scan information—either throughradiologist lesion quantification or through deep learning—to clinical and biologicaldata, improves prediction of severity. These findings show that CT scans containnovel and unique prognostic information, which we included in a 6-variable ScanCovseverity score.
Document type :
Reports
Complete list of metadata

Cited literature [69 references]  Display  Hide  Download
https://hal.archives-ouvertes.fr/hal-02586111
Contributor : Emilie Chouzenoux <>
Submitted on : Friday, July 3, 2020 - 1:24:06 PM
Last modification on : Saturday, May 1, 2021 - 3:49:35 AM

File

Vignette du document
scancovidia_merged.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02586111, version 3

Collections

CEA | FNCLCC | INRIA | CEA-UPSAY | JOLIOT | SHFJ | LARA | CENTRALESUPELEC | CVN | CEA-DRF | IGR | CNRS | COVID19_INRIA | GS-COMPUTER-SCIENCE | GS-ENGINEERING

Citation

Nathalie Lassau, Samy Ammari, Emilie Chouzenoux, Hugo Gortais, Paul Herent, et al.. AI-based multi-modal integration (ScanCov scores) of clinical characteristics, lab tests and chest CTs improves COVID-19 outcome prediction of hospitalized patients. [Research Report] Inria Saclay Ile de France. 2020. ⟨hal-02586111v3⟩

Export

BibTeX TEI DC DCterms EndNote

Share

Metrics

Record views

996

Files downloads

470