University of Twente Student Theses

Login

herQLUS: Assessment of lung aeration by lung ultrasound in intensive care unit patients : development of a quantitative analysis method

Smit, Marry (2018) herQLUS: Assessment of lung aeration by lung ultrasound in intensive care unit patients : development of a quantitative analysis method.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Abstract:Lung Ultrasound (LUS) is an upcoming diagnostic tool in the Intensive Care Unit (ICU) because it can measure lung aeration in ICU patients through different semi-quantitative scoring systems in a non-invasively and bed-side manner. Current LUS scores which are partly based on artifacts can only provide a rough estimation of lung aeration and are observer dependent. The objective of this study was to compare standardized findings on LUS images with corresponding lung regions on the chest CT-scan, and to develop an algorithm that quantifies lung aeration from LUS images. A 12-region protocol LUS exam was performed minutes before a chest CT-scan. The subpleural layers scanned through LUS were localized in the CT-scan where the mean Hounsfield Unit (HU) value was obtained as the reference aeration value. LUS images were scored with the LUS aeration score where regions were scored as; ‘0’, A-pattern; ‘1’, B1-pattern; ‘2’, B2-pattern; or ‘3’, C-pattern. In case of a B-pattern, the conventional B-line score (count of separate B-lines) and the modified B-line score (percentage of screen covered by B-lines) were calculated. LUS scores were compared to HU values from CT with the Pearson correlation coefficient. Finally, an aeration quantification algorithm based on a machine learning algorithm was developed to predict lung aeration from LUS images. The algorithm was verified on artificial LUS images and validated on clinical LUS images. 18 patients were included in this study. A significant correlation was found between the LUS aeration score and the corresponding HU value (R2 = 0.516, p < 0.001). From the different patterns in LUS, the A- and C-pattern corresponds to a small range of HU values. B-patterns showed to correspond to a large range of HU values. The conventional and the modified B-line score compared to the HU values had a correlation coefficient (R2) of 0.004 (p =0.616) and a R2 of 0.466 (p < 0.001) respectively. The developed algorithm verified on artificial LUS images had a Mean Absolute Error (MAE) of 64 HU. When applied to clinical LUS data, the algorithm had an MAE of 196 HU, however the algorithm was able to learn from LUS data. Concluding, the LUS aeration score and the modified B-line score had a good performance where the conventional B-line score performed poor. The developed algorithm was verified on artificial data with a good error rate. On clinical LUS images, the algorithm was also able to predict lung aeration, however the error is large.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/76749
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page