Abstract Archives of the RSNA, 2011
LL-INS-TH1B
A Comparison of Two Methods of Thoracic Soft Tissue Image Generation for Improving Pulmonary Nodule Identification
Scientific Informal (Poster) Presentations
Presented on December 1, 2011
Presented as part of LL-INS-TH: Informatics
Ronald David Novak PhD, Abstract Co-Author: Nothing to Disclose
Elena DuPont, Abstract Co-Author: Nothing to Disclose
Gunhild Erstad Aandal MD, Presenter: Nothing to Disclose
It has been suggested that the suppression of ribs and clavicles in digital chest images has been shown to improve mailgnant nodule detection by Computer Aided Diagnostic (CAD) software. Currently, two FDA approved methods exist for the suppression of calcified tissue in digital chest radiographs. Dual Energy Subtraction Radiography (DESR) and Post-Processing Enhancement (PPE) produced by General Electric Health Systems and Riverain Medical respectively, produce soft tissue images. CAD software was used to identify malignant nodules in both DESR and PPE soft tissue images. Accuracy in malignant nodule identification between the two image types was compared.
A sample of digital postero-anterior (PA) images from 36 subjects with pathology proven pulmonary nodules (9-30 mm) were retrospectively analyzed. DESR produces both normal PA and soft tissue digital images. PPE software (SoftView Ver. 2.1, Riverain Medical) was used to create an additional soft tissue image from the normal digital DESR image. CAD software (OnGuard, Ver. 5.1, Riverain Medical) was used to identify regions of interest (ROI) in each soft tissue image potentially corresponding to a malignant nodule. The number of ROI identified and frequency of inaccurate nodule identification (false positives per image (FPPI) were calculated.
For this version of CAD software, both ROI and FPPI were significantly lower (ROI: 2.2 vs 3.2; p<0.001), (FPPI: 1.4 vs 2.7; p<0.001) in DESR soft tissue images when compared to PPE. However, there was no difference in sensitivity when both DESR and PPE were utilized (SEN: 0.604 vs 0.646; p> 0.10).
In this sample of soft tissue images, a reduction in either ROI and/or FPPI did not necessarily indicate improvement in sensitivity or diagnostic accuracy. However, reductions in both ROI identified and the number of FPPI will encourage the use of CAD for general population screening. Larger sample sizes for further study and improvement in CAD software identification should advance the diagnostic capability of pulmonary nodules by CAD systems.
Many studies will be required to determine the best method for optimum CAD pulmonary malignant nodule identification using soft tissue images.
Novak, R,
DuPont, E,
Aandal, G,
A Comparison of Two Methods of Thoracic Soft Tissue Image Generation for Improving Pulmonary Nodule Identification. Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL.
http://archive.rsna.org/2011/11015986.html