Abstract Archives of the RSNA, 2007
SSG18-07
Computer-aided Detection of Lung Nodules in Low Dose Chest CT: Influence of Image Reconstruction Kernel for CAD Performance
Scientific Papers
Presented on November 27, 2007
Presented as part of SSG18: Chest (Lung Nodules, CAD)
Myung Jin Chung MD, Abstract Co-Author: Nothing to Disclose
Jiyoung Hwang MD, Presenter: Nothing to Disclose
Chin A. Yi MD, PhD, Abstract Co-Author: Nothing to Disclose
Tae Sung Kim MD, Abstract Co-Author: Nothing to Disclose
Younga Bae MD, Abstract Co-Author: Nothing to Disclose
Kyung Min Shin MD, Abstract Co-Author: Nothing to Disclose
Sun Young Jeong, Abstract Co-Author: Nothing to Disclose
Kyung Soo Lee MD, Abstract Co-Author: Nothing to Disclose
et al, Abstract Co-Author: Nothing to Disclose
et al, Abstract Co-Author: Nothing to Disclose
To evaluate the relationship between CT reconstruction kernel and performance of computer aided diagnosis (CAD) system, and to determine which kernel is best for automated lung nodule detection in lung cancer screening CT
We prospectively studied 36 healthy subjects who underwent low dose CT for lung cancer screening. Each subject underwent MDCT scan without contrast enhancement (0.625 mm x 40 detector rows, 120kVp, automatic dose modulation with less than 30mAs/slice). Axial images (1 mm section thickness reconstruction with 1 mm interval) were reconstructed with three different reconstruction kernels; B (standard), C (high), and L (ultra-high spatial frequency algorithm, dedicated kernel for lung CT). All series were reviewed using a commercial CAD system for automatic lung nodule detection. Two radiologists analyzed CT scans in consensus with reference to CAD results, and recorded the presence and size (type A: < 4 mm, type B: 4 ~ 20 mm in diameter) of uncalcified nodules as a standard reference.
The 36 scans showed 230 uncalcified nodules (157 type A and 73 type B nodules). The sensitivities on each series were as; B: 47%, C: 46%, L: 48%. False positive ratios were about 50% (B: 58%, C: 54%, L: 47%). For detection of type B nodules, the sensitivities were higher than 80% (B: 82%, C: 88%, L: 82%). Although overall sensitivity of was best in kernel L, kernel C is best for detection of clinically significant nodules (type B). When the results of two series were joined, the sensitivity was boosted (B+C: 52%, B+L: 60%, C+L: 57% for all nodules, B+C: 89%, B+L: 95%, C+L: 96% for type B).
Sensitivity of CAD system was influenced by selection of reconstruction kernel. In spite the dedicated lung kernel is recommended for visual reading of thin section chest CT, recommendable kernel for CAD system can be different. By union of data from two different kernels, CAD sensitivity can be elevated without further patient exposure.
We must choose adequate kernel to enhance CAD performance. By union of data from two different kernels, CAD sensitivity can be elevated without further patient radiation exposure.
Chung, M,
Hwang, J,
Yi, C,
Kim, T,
Bae, Y,
Shin, K,
Jeong, S,
Lee, K,
et al, ,
et al, ,
Computer-aided Detection of Lung Nodules in Low Dose Chest CT: Influence of Image Reconstruction Kernel for CAD Performance. Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL.
http://archive.rsna.org/2007/5010968.html