Premium
SU‐F‐R‐43: Recursive K‐Means Filter for Preserving Signals of Interest
Author(s) -
Chu A,
Yan P,
Shih R,
Wuu C
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4955814
Subject(s) - hounsfield scale , filter (signal processing) , similarity (geometry) , sorting , cluster (spacecraft) , euclidean distance , computer science , recursive filter , rectangle , mathematics , pattern recognition (psychology) , artificial intelligence , algorithm , computed tomography , computer vision , digital filter , image (mathematics) , medicine , geometry , root raised cosine filter , radiology , programming language
Purpose: To use recursive cluster analysis to filter noises and artifacts out of useful signals. The method was applied to CT or CBCT for preserving interested low Hounsfield Unit (HU) signals such as lung tissues while the HU of lung tissues is often overlapped with CT (or CBCT) artifacts. Methods: Cluster analysis is to search for the similarity among data points, which could employ multi‐dimensional techniques. For our purpose, we searched for a robust, fast, and automatic process for sorting interested low HU object in CT, Therefore, 1‐D cluster analysis is suitable for the purpose. However, the 1‐D cluster analysis might not be able to sensitively separate the groups with overlap in values, for example, lung tissue and low‐HU CT artifacts. We found the recursive k‐mean (initialized by k‐mean++ over Euclidean distance) with division of small size groups can efficiently strip the low noises out and reveal the interested group by a repetitive fashion. Results: The algorithm of repetitively regrouping and filtering with a size of small division can effectively remove the low‐HU values of artifacts. Because artifacts are usually scattered separate groups, the lung tissues buried in the low‐HU groups can be revealed with an appropriate recursive steps. Conclusion: The filtering technique is efficient, robust and can be applied to many different applications as long as the distribution of interested signals can be considered as a cluster in value.