Evaluation of feature-based registration in dynamic infrared imaging for breast cancer diagnosis.

Dynamic infrared imaging has been proposed in literature as an interesting adjunctive technique to mammography in breast cancer diagnosis. For a correct analysis of the infrared image sequences, however, a preprocessing registration step may be fundamental. In this paper, we provide a first quantitative evaluation of the performances of registration on dynamic infrared images for this specific application. We propose an algorithm for the automatic control point localization and employ a feature-based piecewise linear registration approach. The quality of the image alignment is evaluated on a small sample of subjects, considering both alignment and localization errors. Results show that subject movement of up to five pixels is compensated, obtaining a subpixel error correction.