| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 98 === Image-Guided Radiation Therapy (IGRT) is the latest technology to help a therapist decide tumor position more precisely in radiation therapy. It is based on real-time imaging to locate the target during atreatment. However, manual inspection of the images acquired during radiotherapy is labor-intensive and thus highly undesirable. In this thesis, we develop a learning-based algorithm to detect implanted markers automatically using MV images. In the proposed method, we first select marker candidates using the Laplacian of Gaussian (LoG). Then, candidate points are projected onto a low-dimensional subspace determined by Kernel Principal Component Analysis (KPCA) plus Nonparametric Discriminant Analysis (NDA). Finally, we decide whether a candidate point is a marker or not by calculating Mahalanobis distance between projected points. We have conducted a series of experiments on patient images and the results demonstrate the effectiveness of the proposed method.
|
|---|
