Source author record

Manas Saha

Explore this author’s publications, collaborators and research topics. This catalog record has not been claimed by its author.

ResearcherUnclaimed source record

Catalog footprint

Research overview

1works
2topics
4close collaborators

Published work

1 publications

preprint2021arXiv

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

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.

Research graph

See the researcher in context

Open full explorer

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.