Reduction of false positives at vessel bifurcations in computerized detection of lung nodules

Objective: We describe a new false positive (FP) reduction method based on surface features in our computerized detection system for lung nodules and evaluate the method using clinical chest computed tomography (CT) scans. Methods: In our detection method, nodule candidates are extracted using volumetric curvature-based thresholding and region growing. For various sizes of nodules, we adopt multiscale integration based on Hessian eigenvalues. For each nodule candidate, two surface features are calculated to differentiate nodules and FPs at vessel bifurcations. These features are fed into a quadratic classifier based on the Mahalanobis distance ratio. Results: In an experimental study involving 16 chest CT scans, the average number of FPs was reduced from 107.5 to 14.1 per case at 90% sensitivity. Conclusions: This proposed FP reduction method is effective in removing FPs at vessel bifurcations.