Automatic Detection of PET Lesions

We propose a method to handle the automatic detection of PET lesions based on asymmetry feature measurement after segmenting the PET images. It is an essentially initial step in automatic diagnosis and content-based retrieval applications. Our technique includes six steps: Image alignment, segmentation, image reflection & subtraction, thresholding, background removal using morphological filtering and segment back-mapping. Compared with existing per-pixel asymmetry detection methods, our method can provide fewer false positives and more accurate results.