Sang Ok Park MD, Presenter: Nothing to Disclose

Joon Beom Seo MD, PhD, Abstract Co-Author: Nothing to Disclose

Namkug Kim MS, Abstract Co-Author: Nothing to Disclose

Seong Hoon Park MD, Abstract Co-Author: Nothing to Disclose

Young Kyung Lee, Abstract Co-Author: Nothing to Disclose

Bum-Woo Park, Abstract Co-Author: Nothing to Disclose

Yu Sub Sung, Abstract Co-Author: Nothing to Disclose

Youngjoo Lee, Abstract Co-Author: Nothing to Disclose

Jeongjin Lee MS, Abstract Co-Author: Nothing to Disclose

Suk-Ho Kang, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To develop an automated system for quantification of various regional disease patterns at HRCT of diffuse lung diseases.

Six-hundred circular regions of interest (ROI) with 10 pixel diameter, comprising of each 100 ROIs representing six regional disease patterns (normal, NL; ground-glass opacity, GGO; reticular opacity, RO; honeycombing, HC; emphysema, EMPH; and consolidation, CONS) were marked by an experienced radiologist from selected HRCT images. For each ROI, 37 features from texture (histogram, gradient, run-length encoding, and co-occurrence matrix) and shape (cluster and TopHat transform) analyses were extracted. Extracted features were applied to SVM classifier for the training of system. The trained system was applied to 92 HRCT images representing variable diseases and regional disease patterns (UIP, 68; BOOP, 10; emphysema, 6; PCP, 3; pneumonia, 3; BAC, 1; AIP, 1). The whole lung area in each image was assessed and each pixel was classified by using continuous moving ROI function. To validate the quantification result of the system, two radiologists classified the lung area into six patterns using dedicated software independently. The agreements between results of the system and radiologists were evaluated by pixel-based analysis.

The overall accuracy of the system in determining each disease patterns on ROI basis was 89% with 5-fold- cross-validation method. On quantification results, the average agreements between the results of automated system and each radiologists were 52% and 49% respectively (NL, 51%, 41%; GGO, 34%, 53%; RO, 72%, 71%; HC, 67%, 69%; EMPH, 67%, 78%; and CONS, 64%, 55%). The agreement between radiologists was 67% (NL, 67%; GGO, 82%; RO, 56%; HC, 73%; EMPH, 94%; and CONS, 56%). Major source of disagreement was on determining the patterns between NL, GGO and RO.

An automated quantification of regional disease patterns at HRCT is possible with feature-based classification system. Further calibration of the system is needed to improve the performance.

The automated quantification system may be used for objective and reproducible assessment of regional disease severity and interval change of various diffuse lung diseases.

Park, S, Seo, J, Kim, N, Park, S, Lee, Y, Park, B, Sung, Y, Lee, Y, Lee, J, Kang, S, et al, , et al, , Automated Quantification of Regional Disease Patterns at HRCT of Various Diffuse Lung Diseases.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5012324.html