Biomedical Image Classification by Data Reduction

Abstract

The aim of this paper is to consider the quantitative classification of medical images for diagnostic purposes. Medical images present many common features from patient to patient but the differences between them may not always be due to some abnormality. This variety of format for such images leads to a complexity that restricts the success of image processing. This paper considers the use of data compression techniques to provide a compact representation of the image and, hopefully, a more efficient means for classification [1,2]. We have also studied pre-processing (histogram equalisation [2] and spectral filtering) in order to aid establishing decision criteria. We consider chest X-rays of pneumoconiosis patients and acoustic images of babies heads. The only compression techniques discussed here are the discrete cosine transform (DCT) and the Karhunen-Loève transform (KLT), the latter being of more interest. Confidence in finding criteria is drawn from applications of these transforms to simulated data.