There has been tremendous progress in medical imaging of the human body as represented by multi-detector computed tomography (CT), allowing higher scanning speeds and higher spatial resolution than ever before. A human torso is often scanned within a few seconds with 0.5 mm spatial resolution, which results in between several hundred and more than one thousand image sections. Although such CT volume includes rich information for imaging diagnosis, it forces doctors to interpret a large number of axial images. For the purpose of assisting physicians, it is important to develop automated image understanding algorithms of human anatomy.
This chapter presents examples of segmentation algorithms for organs/tissues in the human body, mainly in the trunk. As is well known, by avoiding unnatural segmentation results, a computational anatomy (CA) model that learns statistical variations of an organ/ tissue is very helpful for segmentation. It achieves higher segmentation accuracy than one without a CA model. Chapter 3 focuses on algorithms of medical imaging based on CA models presented in Chap. 2. Since the CT scanner is one of the most common imaging devices in the clinic, this chapter mainly concerns CT data. It should be kept in mind, however, that multimodality medical imaging devices such as PET/CT and PET/MRI are being used more and more. This chapter also describes segmentation algorithms for these. The outline of this chapter is as follows.
Section 3.2 discusses segmentation algorithms of the axial skeleton, i.e., the vertebrae, ribs, followed by hip joints, all of which are the main concern in skeleton segmentation. After reviewing automated segmentation algorithms, a landmark detection framework based on a statistical model for whole spinal anatomical landmarks is introduced and discussed. Subsequently, segmentation and reconstruction algorithms of the hip joint are presented. A skeletal muscle segmentation algorithm from CT volume data is discussed in Sect. 3.3. In Sect. 3.4, segmentation algorithms of lymph nodes in medical images are reviewed followed by a lymph node segmentation algorithm from abdominal CT images. Section 3.5 deals with topics in the brain, head, and neck, and it starts with computational neuroanatomy that is a discipline focused on analyzing and modeling the anatomy of individual brains and the structural variability across a population. In section 3.5.1, voxel-based morphometry (VBM) and deformation-based morphometry (DBM) are introduced. Section 3.5.2 presents algorithms for understanding white matter from diffusion-weighted MR images (DWI) with some clinical applications. A brain CT understanding algorithm based on a normal brain CT model is presented for detection of intracranial hemorrhage in Sect. 3.5.3. Several oral segmentation algorithms using statistical models and algorithms for image understanding of fundus oculi from fundus images and retinal OCT images of retinal layers are also described with their applications to computer-aided diagnosis in Sects. 3.5.4 and 3.5.5. Section 3.6 focuses on thoracic organs, including the tracheobronchial tree, lungs, vessels, and fissures in a thoracic CT volume in the first half of this section. Several segmentation algorithms of these organs followed by an anatomical labeling process are discussed. Since breast ultrasound imaging, mammography, and breast MRI are used in the diagnosis and follow-up of breast cancer, computational model-based segmentation algorithms are presented in Sect. 3.7. In Sect. 3.8, image understanding algorithms of the heart in an echocardiographic image sequence and MR images, and coronary arteries in a CT volume, are provided. Section 3.9 presents CA model-based segmentation algorithms of the abdominal organs, in which a brief survey of segmentation algorithms for each organ is given followed by details of a state-of-the-art algorithm. Multi-organ segmentation schemes based on computational models are also described and discussed in this section. Segmentation algorithms of the abdominal aorta and abdominal vessels are presented followed by anatomical labeling of the segmented vessels.