Template Based Alignment and Interpolation Methods Comparison of Region of Interest in Thermal Images

Abstract

Clinicians often have to compare two or more medical images with each other. Usually it is not the entire image that is important but only particular regions of interest (ROIs). In order to simplify the comparison and to reduce a source of human error a new approach for semiautomatically aligning ROIs is proposed. Using thermal (infrared) images as an example this paper builds on the author's previous work on automatically producing contour outlines of ROIs by applying noise removal and edge detection techniques. This study uses the outcome of the previous work to produce an initial and approximate ROI template shape from one of the images. This template shape is displayed as an overlay onto the second image together with control points which the user can manually move into new positions that match the contours of the second image. Using this shift of the control points the second image can then be transformed (morphed) into the same outline shape as the first image. It is the aim of this work to compare the results of applying three different interpolation methods commonly used for scaling and transforming images. The measures used for comparison are the changes of temperature mean and standard deviation in the ROIs resulting from transformation. Results show that the change in mean temperature is less than 0.5ºC in the worst case and 0.2ºC in the best Nearest Neighbour algorithm).