Page 21 - AIH-2-1

Artificial Intelligence in Health COVID-19 diagnosis: FPA, k-NN, and SVM classifiers

1. Introduction clustering techniques, 24-26 level set techniques, 27,28 graph cut
techniques, 29,30 genetic algorithms, 31,32 artificial intelligence-
The lungs are a pair of spongy, air-filled organs located on based segmentation, 33,34 and hybrid algorithm. In our
35
either side of the chest. Each lung is roughly cone-shaped, work, an optimal thresholding approach has been used to
with its base resting on the diaphragm. The lung has two locate a value acceptable for segmenting the lung CT slice.
1
parts: the right lung, which is larger and has three lobes
(superior, middle, and inferior), and the left lung, which is Computer-aided diagnosis (CAD) systems play an
smaller and divided into superior and inferior lobes. Lung important role in assisting physicians in the process of
2
36
diseases are conditions that obstruct normal lung function. clinical decision-making. In the domain of diagnostic
3
These include a variety of conditions such as chronic radiology, the CAD system is designed to diagnose
obstructive pulmonary disease, pneumonia, asthma, abnormalities in images created by imaging modalities.
acute bronchitis, Coronavirus disease 19 (COVID-19), The imaging modalities are X-rays, CT, high-resolution CT,
pulmonary edema, idiopathic pulmonary fibrosis, positron emission tomography, single-photon emission
sarcoidosis, pleural effusion, pleurisy, bronchiectasis, CT, and magnetic resonance imaging. A CAD system
cystic fibrosis, lymphangioleiomyomatosis, interstitial helps medical professionals by simplifying the process of
lung diseases, lung cancer, tuberculosis, acute respiratory interpreting numerous images created by different types of
distress syndrome (ARDS), and coccidioidomycosis, and imaging, where manual involvement is time-consuming.
37
so on. In this research, early detection of COVID-19 is the In the domain of diagnosing pulmonary disorders, the
4
key focus. CAD system takes the input image obtained from the
COVID-19 is an infectious disease caused by severe imaging modalities, employs computational techniques to
acute respiratory syndrome-coronavirus-2 (SARS-CoV-2), locate suspected abnormalities present in the image, and
which transmits between humans through physical contact, leads to a precise diagnosis. Techniques such as machine
respiratory droplets, and aerosols. The disease is identified learning (ML), image processing, pattern recognition, and
by lung lesions detected through imaging techniques, deep learning (DL) are commonly employed to enhance
such as X-rays and computed tomography (CT) scans. CT abnormality detection in medical images. 38
scans make it easier to assess the presence and severity of In this research, we developed a CAD system to detect
COVID-19 nodules. Moreover, considering the structural the presence or absence of COVID-19. First, an adaptive
or anatomical details of the lung that are essential for Wiener filter was used to eliminate the additive noises.
the detection analysis, CT imaging outperforms X-ray Then, optimal thresholding was used to segment the
5
radiography in providing knowledge on these. The typical lungs, and relevant features were extracted. To select the
signs of lung lesions, such as ground glass opacity (GGO) optimal feature set, a bio-inspired wrapper-based flower
in the early stages and consolidation in the later stages, pollination technique was employed, using the accuracy
could be observed from CT slices. Studies have reported of the k-NN classifier as the fitness function. The support
6
that radiological imaging, such as CT and X-rays, may be vector machine (SVM) classifier was then trained using the
helpful in supporting the early screening of COVID-19. selected optimal subset of features.
7,8
Although real-time polymerase chain reaction (RT-PCR)
is considered the gold standard for diagnosing COVID-19, This framework can be generalized for applications
recent advancements in medical imaging have significantly in biomedical lung imaging diagnosis. This manuscript
improved the diagnosis and quantification of various is structured as follows: Section 2 discusses the relevant
diseases. Using RT-PCR results as a reference, a study of literature; Section 3 outlines the system’s methodology;
1,014 patients in Wuhan, China, achieved an accuracy of Section 4 summarizes the dataset, compares classifiers,
0.68, a sensitivity of 0.97, and a specificity of 0.25 for CT evaluates other state-of-the-art approaches, and presents
slices indicating COVID-19 infection. 9 the experimental findings; Section 5 offers conclusions and
recommendations for future work.
Segmentation is the process of partitioning lung tissues
with accurate boundaries from CT slices by eliminating 2. Related works
surrounding anatomical structures, such as bones and fat
tissues. The objective of segmentation is to extract regions 2.1. Segmentation techniques for CAD to detect
10
of interest (ROIs) within the lung region to differentiate COVID-19
abnormality from anatomical background. There are 10 Segmentation is an essential step in image processing
different segmentation techniques for lung imaging, and analysis for the assessment and quantification of
11
including: thresholding, 12-15 region growing method, 16-18 COVID-19. It delineates the ROIs, namely, lung, lobes,
watershed algorithm, 19-21 active contour model, 22,23 lesions or infected regions, and bronchopulmonary

Volume 2 Issue 1 (2025) 15 doi: 10.36922/aih.3349