Discriminant Functions and Multi-Resolution Analysis (MRA) for Disease Detection

Problems associated with the detection of diseases in their early stage are well known when using chest radiograph images. A graphical method involving wavelet coefficients as the feature vector (WFV) has been proposed for the detection and discrimination of Mycobacterium Tuberculosis (MTB) and lung cancer (LC). In a pilot study confirmed cases showing no complications (for example a section of the lung filled with water) were studied. Further, in the pilot study the feature vector were compressed for simpler data management. However discrimination using the compressed WFV was developed to handle small sample situations. In this paper, an alternative method for larger samples sizes was employed. A control group was used to calculate the parameters of the Linear Discriminant Function (LDF( x )) and the Quadratic Discriminant Function, (QDF( x )). A separate test group was then used to calculate misclassification probabilities. The feature vector selected, that is vector x , were the average and detail vectors from the MRA of the WFV. The technique developed here allows misclassified cases to be reclassified correctly, a facility not provided for in earlier techniques. Key-Words:Chest Radiograph, Mycobacterium Tuberculosis, Lung Cancer, Wavelet, Multi-Resolution Analysis, Discriminant Functions, Misclassification Probabilities.