COVID-19 Diagnosis and Segmentation Using Machine Learning Analyses of Lung Computerized Tomography Page: 81
This dissertation is part of the collection entitled: UNT Theses and Dissertations and was provided to UNT Digital Library by the UNT Libraries.
Extracted Text
The following text was automatically extracted from the image on this page using optical character recognition software:
Table 5.3: Dataset 3: SemiSeg dataset [50]
Images 1190 170 340
Patients 62 9 17
5.4 Pre-Processing
Pre-processing is done to prepare the datasets. First, the volume data with .nii files is
converted to slice level .jpeg using the MIPAV tool [96] and the NIfTI-Image-Converter [77]. The
dimensions of the original volume data are NxCxHxWxD, where N is the number of images, C
denotes the channels, H is the height of the image, W is the width of the image and D is the depth
corresponding to all image slices for a single CT. After this conversion, the slice level dimensions
are NxCxHxW. Note that there is a single channel for the binary mask and the red, blue, and
green channels for a CT image.
Then, the images and their corresponding lung mask images are loaded in batches. The
images are rescaled to the fixed pixel spacing of 2,2, the image intensities are clipped to the fixed
window and the input images were normalized to have pixel values ranging from (0, 1). Non-
COVID-19 images that get tagged as 'negative' should ideally be excluded. Note that all datasets
we used only had COVID-19 positive patients' data. Some slices show signs of COVID-19, but the
physician-provided mask has all black pixels because not every slice of a positive patient will
manifest disease. Those frames are the 'positive without a mask showing the COVID-19 positive
class' frames and those frames which have the COVID-19 diseased areas are 'positive with a mask
showing the COVID-19 positive class.' Both these categories will be included in training,
validation, and test sets to mimic real-world data. While some published works discarded the
frames with zero annotations i.e., blank annotations to make the model train better, we chose81
Upcoming Pages
Here’s what’s next.
Search Inside
This dissertation can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Dissertation.
Mittal, Bhuvan. COVID-19 Diagnosis and Segmentation Using Machine Learning Analyses of Lung Computerized Tomography, dissertation, August 2021; Denton, Texas. (https://digital.library.unt.edu/ark:/67531/metadc1833510/m1/95/: accessed May 27, 2024), University of North Texas Libraries, UNT Digital Library, https://digital.library.unt.edu; .