Position Prediction in CT Volume Scans
Franz Graf, Hans-Peter Kriegel, Sebastian Pölsterl, Matthias Schubert
Published at ICML 2011 Workshop on Machine Learning for Global Challenges
Conference Date: June 28th - July 3rd, 2011
Conference Location: Bellvue Settle, WA, USA.
Abstract:
In their daily work, radiologists often need to localize and align parts of CT volume scans to perform among others differential diagnosis. In these cases, it is desired to load only the relevant sub volume of the scan and to align the sub volume automatically to the correct position of the query scan. Common techniques employ landmark detectors and interpolation to solve this problem. Yet, these techniques are are not applicable in case of very small volume scans where the query scan comprises only a small amount of images. In this paper, we propose a method to use small sub volumes in CT volume scans for identifying and aligning CT Scans. Our solution employs combinations of weighted image descriptors and instance-based regression and thus demonstrates the need for machine learning techniques in the case of position prediction. The experiments show that the new method improves the mean error and standard deviation by 6% and 10%, respectively, compared to a state of the art method.
Documents:
This is the author’s version of the work. The original publication is available at the Workshop Website.
(same as the BibTex
@INPROCEEDINGS{GraKriPoeSchetal11,
AUTHOR = {F. Graf and H.-P. Kriegel and S. Pölsterl and M. Schubert and A. Cavallaro},
TITLE = {Position Prediction in CT Volume Scans},
BOOKTITLE = {Proceedings of the 28th International Conference on Machine Learning (ICML) Workshop on Learning for Global Challenges, Bellevue, Washington, WA},
YEAR = {2011}
}
