Comparing Data Strategies for COVID-19 Segmentation in CT Scans
Author Information
Author(s): Oliveira Anne de Souza, Costa Marly Guimarães Fernandes, Costa João Pedro Guimarães Fernandes, Costa Filho Cícero Ferreira Fernandes
Primary Institution: Federal University of Amazonas
Hypothesis
How do different data partitioning strategies affect the segmentation of COVID-19 areas in CT scans using convolutional neural networks?
Conclusion
The slice strategy for segmenting COVID-19 areas in CT scans performed slightly better than the CT-scan strategy.
Supporting Evidence
- Automatic lung segmentation achieved a Dice metric of 84.66% for the slice strategy.
- Manual lung segmentation achieved a Dice metric of 85.35% for the slice strategy.
- The differences in performance metrics between the two strategies were statistically significant.
Takeaway
This study looked at how to better find COVID-19 in CT scans using computers, and found one way worked a little better than another.
Methodology
The study used two datasets and two deep neural network architectures to evaluate the automatic segmentation of lungs and COVID-19 areas.
Potential Biases
The study may have bias due to the small sample size and the reliance on specific datasets.
Limitations
The study had a small number of common images in the two datasets, limiting comparisons.
Participant Demographics
The datasets included CT scans from patients with COVID-19, but specific demographics were not detailed.
Statistical Information
P-Value
0.00033
Confidence Interval
[66.823–79.197], [82.161–87.159]
Statistical Significance
p<0.01
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website