Evaluation of reproducibility for manual and semi-automated feature extraction in CT and MR images

Three methods for extraction and quantitative measurement of features in CT and MR images are examined: hand tracing, semi-automated tracing using the livewire graph search algorithm, and extraction using a geometrically constrained region growth algorithm. Extracted structures are evaluated in terms of volume, cross-sectional area, and major axis in plane. Reproducibility, time required for extraction, and accuracy of each of these metrics is measured for each of the extraction methods using both phantoms and clinical lung tumor data.