French
English
- Participants
J.C. Richard, M. Orkisz, A. Morales Pinzón, L. Bitker
- National and international collaborations
Project C15M04, programme ECOS Nord :
M. Hernández Hoyos (Univecrsidad de los Andes, Bogotá, Colombia)
L. Flórez Valencia (Pontificia Universidad Javeriana, Bogotá, Colombia)
Project 34852WG, PHC Polonium :
A. Fabijańska (Politechnika Łódzka, Poland)
- Question
Is it possible to impact ARDS prognostic using computed tomography to adapt mechanical ventilation to patients’ individual profile?
- Objective
To evaluate alveolar recruitment and cyclic recruitment-derecruitment in several ventilatory conditions during experimental ARDS and in ARDS patients. To determine optimal ventilatory settings using CT. To evaluate the impact on ARDS prognostic of a CT-based strategy to adapt mechanical ventilation in ARDS patients, in comparison with a control group.
- Methodology
CT image segmentation: parenchyma [bib]MORA-17b[/bib], airways [bib]LIDA-17[/bib], vessels [bib]ORKI-14[/bib], [bib]RUDY-14[/bib]. Cascade and hybrid registration [bib]GOME-16[/bib]: grey levels and graph match [bib]MORA-17a[/bib]. New definition of recruitment at voxel scale [bib]ORKI-19[/bib].
The main methodological lock is the development of an algorithm able to segment the lung in “real time” (30 min to 1 hour).
- Results and illustrations
Sensitivity of 94,3% and precision of 92,8% in matching graphs representing the bronchial tree segmented in images acquired at different end-expiratory pressures [bib]MORA-17a[/bib].
Figure 1. Bronchial trees segmented at different end-expiratory pressures.
Overlap (Dice coefficient) amounting to 96% between lung automatic segmentation, obtained by an auto-atlas method, and reference segmentation by manual tracing on the whole lung. However, overlap was reduced, with a Dice coefficient amounting to 81%, in the postero-basal non-aerated parts of ARDS lungs lacking contrast between parenchyma and surrounding tissue [bib]MORA-17b[/bib].
Figure 2. 3D lung segmentation with automatic method (red, middle and right), and with manual tracing using interactive software (left), superimposed with fully manual reference tracing (green)
Possibility of displaying local-recruitment maps at voxel scale, based on 3D image registration, and good consistency (coefficient of determination greater than 0,95) between regional recruitment and local-recruitment values integrated over the considered region [bib]ORKI-19[/bib].
Figure 3. Intratidal local-recruitment map superimposed onto the original CT image of a subjet with experimental ARDS.
- Grants
Bernard Dräger award of the European Society of Intensive Care Medicine - ECCRN 2010 : 15000 Euros.
- Other publications of the team
- Medical: [bib]RICH-14[/bib], [bib]GIRA-14[/bib], [bib]HERN-13[/bib], [bib]POUZ-13[/bib], [bib]WALL-13[/bib], [bib]RICH-13[/bib], [bib]GUER-13b[/bib], [bib]GUER-13a[/bib], [bib]RICH-13e[/bib], [bib]RICH-13d[/bib], [bib]RICH-13c[/bib], [bib]RICH-13b[/bib], [bib]RICH-13a[/bib].
- On image processing: [bib]MORA-16b[/bib], [bib]SOLO-16[/bib], [bib]FLOR-15[/bib], [bib]MORA-14c[/bib], [bib]MORA-14b[/bib], [bib]MORA-14a[/bib].