EFFECT OF IMAGE PRE-PROCESSING METHOD ON CONVOLUTIONAL NEURAL NETWORK CLASSIFICATION OF COVID-19 CT SCAN IMAGES

Khanun Roisatul Ummah, Tita Karlita*, Riyanto Sigit, Eko Mulyanto Yuniarno, I. Ketut Eddy Purnama, Mauridhi Hery Purnomo

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

6 Citations (Scopus)

Abstract

The presence of abnormalities on CT lung images of COVID-19 patients, such as ground-glass opacity and consolidation, can be used to aid in the detection of COVID-19. These abnormalities can be ambiguous and obscure, resulting in false detection by the doctor. This paper evaluates several image pre-processing methods for automatic detection of COVID-19 based on CT images. First, we used Watershed segmentation to separate the lung cavities. We retained the interior of the lung cavity, where the features of COVID-19 were located. Next, we evaluate the addition of a smooth-ing image method of Median and Gaussian filters to remove blood vessel spots. We also assess the contrast improvement-based methods using Histogram Equalization (HE) and Contrast Limited Adaptive Histogram Equalization (CLAHE) to highlight the features of COVID-19 further. Multi-dimensional extension of CLAHE (MCLAHE) is also used as an optimization of CLAHE method. In addition, the Inverted Threshold to Zero method is utilized to segment the COVID-19 features. We used transfer learning Convolution-al Neural Network (CNN) in VGG-19, ResNet50, Xception, and DenseNet201 for the classification process. The results show that classification accuracy can be improved by adding appropriate pre-processing techniques. CLAHE and MCLAHE have the highest accuracy with 91.20% and 91.60%, respectively.

Original languageEnglish
Pages (from-to)1895-1912
Number of pages18
JournalInternational Journal of Innovative Computing, Information and Control
Volume18
Issue number6
DOIs
Publication statusPublished - Dec 2022
Externally publishedYes

Keywords

  • CNN
  • COVID-19 lassifiation
  • CT san
  • Image pre-proessing
  • Transfer learning

Fingerprint

Dive into the research topics of 'EFFECT OF IMAGE PRE-PROCESSING METHOD ON CONVOLUTIONAL NEURAL NETWORK CLASSIFICATION OF COVID-19 CT SCAN IMAGES'. Together they form a unique fingerprint.

Cite this