Multi-modal medical image segmentation based on vector-valued active contour models

Abstract

Positron emission tomography (PET), magnetic resonance imaging (MRI) and computed tomography (CT) are widely utilized medical imaging modalities that provide essential anatomic and structural details. Many medical segmentation methods are not effective for a single-modal image of poor quality (e.g., low contrast in CT or low spatial resolution in PET). For practical radiotherapy treatment planning, multi-modal imaging information is regularly used. In this paper, a novel vector-valued active contour model is proposed to segment multi-modal medical images simultaneously for abnormal tissue regions. The method makes use of the functionality information and anatomical structure information advantages from each modality. Since each modality has its own signal characteristics, we use region-based information, combining hybrid mean intensities simultaneously. Furthermore, by utilizing a two-dimensional vector field with different image modalities, edge-based information is used to constrain the results of the image segmentation. The proposed approach is evaluated on datasets including lung PET-CT and brain MRI-CT images. Our qualitative and quantitative research results confirm the effectiveness of the proposed method.