Measurement of Local Deformation due to Lung Tumor Response to Radiation Therapy

Accurate alignment of intra-subject CT acquired during radiotherapy is challenging due to tumor regression introducing large image content differences. Some regressing tumors tend to expose newly-visible lung parenchyma, in a type of regression we term ‘infiltrative’. We evaluated two nonrigid registration algorithms for accuracy and plausibility in aligning two images demonstrating infiltrative regression. Both used a vesselness measure-based feature cost and a linear elastic penalty term. The performance of sum of square difference and sum of square tissue volume difference intensity costs were compared in three subjects. Landmark error was not significantly different between the two algorithms, although sum of square tissue volume difference produced more plausible transformations in the infiltrative regression scenario.