Page 22 - AIH-2-1

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

segments, in the chest X-ray or CT slices. Segmented regions transforms, were extracted using the method of principal
could be further used to extract features for diagnosis and component analysis. In the classification, k-NN, sparse
other applications. This subsection summarizes the related representation classifier (SRC), artificial neural network
segmentation works in COVID-19. (ANN), and SVM classifiers were used for normal,
pneumonia, and COVID-19 classifications. Nine different
Khin et al. have proposed a segmentation algorithm
39
to detect COVID-19 in chest CT slices using Deeplab datasets collected from various sources were examined.
The accuracies achieved were 91.70%, 94.40%, 96.16%, and
v3 . The dataset used was the COVID-19 radiography 99.14% by k-NN, SRC, ANN, and SVM, respectively, for
+
database, which contains a total of 15,153 images, COVID-19 diagnosis.
including 10,192 normal images, 3,616 COVID-19 images,
44
and 1,345 pneumonia images. Since the dataset was highly Shankar et al. have suggested a CAD system for
imbalanced, five different approaches were employed. The diagnosing COVID-19 using chest X-ray images. Initially,
ensemble of convolutional neural network (CNN) with the Wiener filter was used to pre-process images. The
image augmentation achieved an accuracy of 99.23%. fusion-based feature extraction method was subsequently
carried out using GLCM, gray level run length matrix,
Venkatesan et al. have introduced an automated image and local binary patterns. The ideal feature subset was
40
processing scheme to extract the COVID-19 lesions from then determined using the Salp swarm algorithm. The
lung CT scan images. In their work, the firefly algorithm images were divided as infected or healthy using an ANN.
and Shannon Entropy-based multi-threshold were used The obtained outcomes outperformed state-of-the-art
to enhance the pneumonia lesions, followed by Markov- techniques. The proposed CAD model’s experimental
Random-Field segmentation to extract the lesions with results showed 95.1% and 95.65% accuracy for binary and
better accuracy. The dataset was obtained from the multiple classes, respectively.
COVID-19 database, which includes 100 images for training
45
and 45 images for testing. The proposed scheme was tested Kadry et al. have proposed a classification technique
and validated using a class of COVID-19 CT images, using a machine learning system (MLS) to classify the
achieving a mean accuracy >92% in lesion segmentation. CT slices as healthy or affected by COVID-19. The
MLS includes five steps, namely, tri-level thresholding,
Chandra has demonstrated a segmentation approach segmentation of the image, feature extraction, feature
41
using the Cuckoo search algorithm with Otsu’s image ranking, implementation of serial fusion, and classifier
thresholding for the extraction of COVID-19 pneumonia implementation and validation. This proposed system was
infection. The proposed approach used Otsu’s/Kapur to tested with 500 images, which includes 250 normal and
enhance the value with a threshold of three and employed 250 COVID-19-affected images obtained from benchmark
Level Set techniques to extract ROIs. The dataset included datasets (Table 1). The proposed MLS achieved an accuracy
COVID-19 images from 20 patients, and the approach of 89.80%.
achieved a segmentation accuracy of 97.62.
2.2. CAD system to detect COVID-19 using
Mohammed et al. have proposed a CAD system for supervised and un-supervised techniques
42
the diagnosis of COVID-19 disease from chest X-ray
46
images. This system can be used to differentiate COVID-19 Wu et al. have proposed a classification system using
from other viral pneumonia-like Middle East respiratory a random forest (RF) classifier for the diagnosis of
syndrome, SARS, and ARDS. Segmentation was performed COVID-19 disease. The dataset description has been given
using Li’s method, followed by the application of Law’s in Table 1. In the proposed system, 11 key features were
42
42
masks to enhance secondary details in the segmented chest selected from 49 features. The model was trained with 11
images. Texture features were then extracted using the key features and achieved an accuracy of 96.95%.
gray-level co-occurrence matrix (GLCM). The obtained Banerjee et al. have suggested a binary classification
47
feature vectors were used to build SVM ensemble models. model utilizing ANN, Logistic regression (LR), and LASSO
Then, the choices of ensemble classifiers were put together Elastic Net Regularized Generalized Linear Models. The
using a weighted voting method. The proposed CAD dataset comprised 598 full blood count results obtained
system achieved an accuracy of 98.04%. from COVID-19 patients. The model with LR achieved an
Bhargava et al. have introduced an automatic accuracy of 87% for the diagnosis of COVID-19 disease.
43
detection system for the diagnosis of COVID-19 from Moutaz et al. have demonstrated an artificial
48
CXR and chest CT slices. Segmentation was done using the intelligence technique based on deep CNN to detect
FCM algorithm. Four types of features, namely, histograms COVID-19 disease. The dataset was obtained from the
of gradients, textural, statistical, and discrete wavelet Kaggle dataset, which has 128 images, including 28 healthy

Volume 2 Issue 1 (2025) 16 doi: 10.36922/aih.3349