Semi-automatic construction of reference standards for evaluation of image registration.

Quantitative evaluation of image registration algorithms is a difficult and under-addressed issue due to the lack of a reference standard in most registration problems. In this work a method is presented whereby detailed reference standard data may be constructed in an efficient semi-automatic fashion. A well-distributed set of n landmarks is detected fully automatically in one scan of a pair to be registered. Using a custom-designed interface, observers define corresponding anatomic locations in the second scan for a specified subset of s of these landmarks. The remaining n-s landmarks are matched fully automatically by a thin-plate-spline based system using the s manual landmark correspondences to model the relationship between the scans. The method is applied to 47 pairs of temporal thoracic CT scans, three pairs of brain MR scans and five thoracic CT datasets with synthetic deformations. Interobserver differences are used to demonstrate the accuracy of the matched points. The utility of the reference standard data as a tool in evaluating registration is shown by the comparison of six sets of registration results on the 47 pairs of thoracic CT data.

Lung nodule volumetry: segmentation algorithms within the same software package cannot be used interchangeably.

OBJECTIVE: We examined the reproducibility of lung nodule volumetry software that offers three different volumetry algorithms. METHODS: In a lung cancer screening trial, 188 baseline nodules >5 mm were identified. Including follow-ups, these nodules formed a study-set of 545 nodules. Nodules were independently double read by two readers using commercially available volumetry software. The software offers readers three different analysing algorithms. We compared the inter-observer variability of nodule volumetry when the readers used the same and different algorithms. RESULTS: Both readers were able to correctly segment and measure 72% of nodules. In 80% of these cases, the readers chose the same algorithm. When readers used the same algorithm, exactly the same volume was measured in 50% of readings and a difference of >25% was observed in 4%. When the readers used different algorithms, 83% of measurements showed a difference of >25%. CONCLUSION: Modern volumetric software failed to correctly segment a high number of screen detected nodules. While choosing a different algorithm can yield better segmentation of a lung nodule, reproducibility of volumetric measurements deteriorates substantially when different algorithms were used. It is crucial even in the same software package to choose identical parameters for follow-up.

A large-scale evaluation of automatic pulmonary nodule detection in chest CT using local image features and k-nearest-neighbour classification.

A scheme for the automatic detection of nodules in thoracic computed tomography scans is presented and extensively evaluated. The algorithm uses the local image features of shape index and curvedness in order to detect candidate structures in the lung volume and applies two successive k-nearest-neighbour classifiers in the reduction of false-positives. The nodule detection system is trained and tested on three databases extracted from a large-scale experimental screening study. The databases are constructed in order to evaluate the algorithm on both randomly chosen screening data as well as data containing higher proportions of nodules requiring follow-up. The system results are extensively evaluated including performance measurements on specific nodule types and sizes within the databases and on lesions which later proved to be malignant. In a random selection of 813 scans from the screening study a sensitivity of 80% with an average 4.2 false-positives per scan is achieved. The detection results presented are a realistic measure of a CAD system performance in a low-dose screening study which includes a diverse array of nodules of many varying sizes, types and textures.