Abstract:

Part 1: Computer-aided detection of interstitial abnormalities in chest radiograph;

The purpose was to develop a computer-aided detection (CAD) system to locate diffuse parenchymal lesions in chest radiographs and compare its performance to that of human experts. The system analyzes the complete lung fields using a two-class supervised pattern classification approach to distinguish between normal texture and texture affected by interstitial lung disease. Analysis is done pixel-wise and produces a probability map for an image where each pixel in the lung fields is assigned a probability of being abnormal. A new, semi-automatic method is proposed for setting a reference standard for training and

evaluating the CAD system. The proposed method employs the fact that interstitial lesions are more distinct on a computed tomography (CT) scan than on a radiograph. The system performance is evaluated with ROC analysis, with separate ROC curves

built for classification of abnormalities of different degrees of subtlety versus normal class. The best performance is achieved

for the linear discriminant and support vector machine classifiers, with an area under ROC curve 0.90 for discrimination between obviously abnormal and normal pixels, and 0.78 for differentiation between any abnormal and normal pixels. The system performance does not significantly differ from that of the observers, when the perihilar regions (around mediastinum) are excluded from evaluation.

Part 2: Application of computerized texture analysis of CT lung images for estimation of interstitial lung disease progression"

The purpose was to develop an automated system for quantification and assessment of interval changes in interstitial lung disease (ILD) based on amounts of certain patterns of abnormal textures on high-resolution CT (HRCT) scans. The system analyzes the pairs of corresponding 2D axial slices of a baseline and follow-up CT scan. Scan pairs are aligned using a non-rigid registration method. In the lung fields regions of the hyperlucency, fibrosis, ground glass, solid and focal textures are automatically identified using a dedicated texture classification algorithm. Differences in the relative amounts of each texture type are used as features to train and test a k-nearest neighbor classifier which assigned a disease progression label for each pair of HRCT sections. As a reference standard the consensus scores of two expert radiologists were used that classified the extent of a disease into “regression”, “stable” or “progression” categories. From 5 random divisions of pairs into training and test sets (40% and 60%) an average accuracy was computed. The automated classifier yielded an average accuracy of 0.75 (standard deviation 0.02). 96% of misclassifications occurred between stable disease and progression or regression.  

About the Speaker: