A relook at logistic regression methods for the initial detection of lung ailments using clinical data and chest radiography

  • Authors:

  • Omar Mohd Rijal;Mohd. Iqbal;Ashari Yunus;Norliza Mohd. Noor

  • Affiliations:

  • Institute of Mathematical Science, Faculty of Science, University Malaysia;Institute of Mathematical Science, Faculty of Science, University Malaysia;Institute of Respiratory Medicine, Kuala Lumpur, Malaysia;Dept. of Electrical Engineering, College of Science and Technology, Universiti Teknologi Malaysia

  • Venue:
  • WSEAS Transactions on Information Science and Applications
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

The problem of diagnosing patients with lung ailments such as Tuberculosis (PTB), Pneumonia (PNEU) and Lung Cancer (LC) when making their initial visit to a medical institution is the focus of this study. Clinical data involving symptoms and signs are used to make important decisions before the availability of the results of further tests. In practice, Logistic Regression Methods are frequently involved in this type of decision making. However, the problem of missing values when the numerical values of certain explanatory variables are not available persists in practical situations. In this paper a logistic regression model using four variables (age, cough, loss of weight (LOW) and loss of appetite (LOA)) are investigated for each of the three diseases. The main result of this study is that the probability of misclassifying the three disease type is large, and that good model fitting does not guarantee correct diagnosis. As a viable substitute, a graphical method of detection with an 85% chance of correct classification based on information extracted from the chest radiograph images is proposed.