We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

eess.IV

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Electrical Engineering and Systems Science > Image and Video Processing

Title: Potential Features of ICU Admission in X-ray Images of COVID-19 Patients

Abstract: X-ray images may present non-trivial features with predictive information of patients that develop severe symptoms of COVID-19. If true, this hypothesis may have practical value in allocating resources to particular patients while using a relatively inexpensive imaging technique. The difficulty of testing such a hypothesis comes from the need for large sets of labelled data, which need to be well-annotated and should contemplate the post-imaging severity outcome. This paper presents an original methodology for extracting semantic features that correlate to severity from a data set with patient ICU admission labels through interpretable models. The methodology employs a neural network trained to recognise lung pathologies to extract the semantic features, which are then analysed with low-complexity models to limit overfitting while increasing interpretability. This analysis points out that only a few features explain most of the variance between patients that developed severe symptoms. When applied to an unrelated larger data set with pathology-related clinical notes, the method has shown to be capable of selecting images for the learned features, which could translate some information about their common locations in the lung. Besides attesting separability on patients that eventually develop severe symptoms, the proposed methods represent a statistical approach highlighting the importance of features related to ICU admission that may have been only qualitatively reported. While handling limited data sets, notable methodological aspects are adopted, such as presenting a state-of-the-art lung segmentation network and the use of low-complexity models to avoid overfitting. The code for methodology and experiments is also available.
Subjects: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2009.12597 [eess.IV]
  (or arXiv:2009.12597v2 [eess.IV] for this version)

Submission history

From: Anwaar Ulhaq Dr [view email]
[v1] Sat, 26 Sep 2020 13:48:39 GMT (2362kb,D)
[v2] Thu, 21 Jan 2021 12:43:04 GMT (2324kb,D)

Link back to: arXiv, form interface, contact.