Ensemble-Based Classification of COVID-19 Radiographic Images Using Spatial, Textural, and Intensity-Based Features

EasyChair Preprint no. 11659

Abstract

The COVID-19 pandemic has underscored the critical need for precise diagnostic tools. Radiographic imaging, particularly chest X-rays and CT scans, has emerged as a vital modality for detecting COVID-19-related lung abnormalities. Several machine learning techniques exist to diagnose Covid-19 from such imaging modalities. In this article, an innovative approach is introduced using multimodal features extracted from CT scan images and ensemble learning to enhance the COVID-19 diagnosis.   The dataset used here encompasses a wide range of COVID-19-positive and non-COVID-19 cases. To improve image quality and information extraction, preprocessing techniques like resizing, Gaussian blur, and dilation are employed. These steps enhance the ability to capture informative image features.  Intensity-based, spatial, and textural features have been exploited for enhanced classification. Intensity-based features capture statistical properties of pixel intensities, while spatial features quantify geometric characteristics of image regions, and texture features leverage the Gray-Level Co-occurrence Matrix (GLCM) to encompass essential texture attributes of the images. The classification strategy used I this paper revolves around ensemble learning, utilizing three diverse base classifiers: Random Forest, Support Vector Machine, and Light Gradient Boosting Machine, each with unique strengths in image classification. These base classifiers are trained on standardized feature vectors from the training dataset, and their predictions are aggregated through averaging the decisions. This ensemble approach yields a robust model for COVID-19 image classification, achieving an overall test accuracy of 92.42%. The performance of the ensemble model is assessed using a dedicated test dataset, demonstrating its superiority (in terms of accuracy) over other methods that utilize hand crafted as well as deep features.

Keyphrases: ensemble learning, Gray Level Co-occurrence Matrix, Light GBM Classifier., Random Forest Classifier, Support Vector Classifier (SVC)

@Booklet{EasyChair:11659,
  author = {Susmita Ghosh and Abhiroop Chatterjee},
  title = {Ensemble-Based Classification of COVID-19 Radiographic Images Using Spatial, Textural, and Intensity-Based Features},
  howpublished = {EasyChair Preprint no. 11659},

  year = {EasyChair, 2024}}