Computerized analysis of interstitial lung diseases on chest radiographs based on lung texture, geometric-pattern features and artificial neural networks

For computerized detection of interstitial lung disease on chest radiographs. we developed three different methods: texture analysis based on the Fourier transform. geometric-pattern feature analysis, and artificial neural network (ANN) analysis of image data. With these computer-aided diagnostic methods. quantitative measures can be obtained. To improve the diagnostic accuracy. we investigated combined classification schemes by using the results obtained with the three methods for distinction between normal and abnormal chest radiographs with interstitial opacities. The sensitivities of texture analysis, geometric analysis and ANN analysis were 88.0+/-1.6%, 91.0+/-2.6% and 87.5+/-1.9%, respectively, at a specificity of 90.0%. whereas the sensitivity of a combined classification scheme with tie logical OR operation was improved to 97.1+/-1.5% at die same specificity of 90.0%. The combined scheme can achieve higher accuracy than the individual methods for distinction between normal and abnormal cases with interstitial opacities.