Abstract | ||
---|---|---|
This paper describes a method for detecting small nodules from 3D X-ray CT images, considering small nodules with CT values of at least -600 H.U. and diameters of at least 2 mm. The proposed method is largely composed of two stages. In the first stage, for the region obtained by the threshold processing, the shape feature parameters of the figure calculated from the distance values in the connected component are applied, and the initial nodule candidate regions are extracted by discriminating the nodule region and the vessel/bronchus region. In the second stage, a minimum directional difference filter in which the filter radius is adaptively adjusted according to the size of the initial nodule candidate region is applied in order to reduce the overextracted regions. The proposed method is applied to seven actual chest CT images (six pathological cases, including multiple nodules, and one normal case), and the result for 361 nodules with a CT value not less than -600 H.U. and diameter not less than 2 mm is that the number of overextracted regions per case is 7.4 on the average, when the detection rate is set as 71%. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(9): 55–64, 2005; Published online in Wiley InterScience (). DOI 10.1002/scj.20175 |
Year | DOI | Venue |
---|---|---|
2005 | 10.1002/scj.v36:9 | Systems and Computers in Japan |
Field | DocType | Volume |
Computer vision,X-ray,Artificial intelligence,Connected component,Mathematics | Journal | 36 |
Issue | Citations | PageRank |
9 | 0 | 0.34 |
References | Authors | |
1 | 7 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yosuke Hayase | 1 | 0 | 0.34 |
Yoshito Mekada | 2 | 115 | 16.08 |
Kensaku Mori | 3 | 1125 | 160.28 |
Jun-ichi Hasegawa | 4 | 221 | 61.17 |
Jun-ichiro Toriwaki | 5 | 578 | 136.04 |
Masaki Mori | 6 | 144 | 17.48 |
Hiroshi Natori | 7 | 220 | 28.49 |