Exploring the Effects of Compression via Principal Components Analysis on X-ray Image Classification

Abstract

Image compression in medical applications implores careful consideration of the effects on data veracity. The inexorable challenge of assessing the volume-veracity trade-off is becoming more prevalent in this critical application area, and particularly when machine learning is used for the purpose of assisted diagnostics. This paper investigates the impact of compressing X-ray images on the accuracy of fracture diagnostics. The accuracy of the classification system is assessed for X-ray images of both healthy and fracture bones when subjected to different levels of compression. Compression is achieved using principal components analysis. Results indicate that accuracy is only marginally affected under a level one compression but begins to deteriorate under level two compression. These results are potentially useful as the level one compression yields gains up to 94% with less than a 2% drop in classification accuracy.