Whole-Body Computational Anatomy

Impact of Whole-Body Imaging

Yoshinobu Sato

Graduate School of Information Science, Nara Institute of Science and Technology, Ikoma-shi, Nara, 630-0192, Japan e-mail: This email address is being protected from spam bots, you need Javascript enabled to view it

This section discusses how whole-body imaging has influenced radiological diagnosis, medical image analysis, medical education, and related research fields.

As a result of rapid improvement in scanning speeds and resolution of MDCT, whole-body (or full-body) imaging is now widely available in clinical practice. CT imaging provides high-resolution image data which precisely delineate the anatomical structures of the whole body, and it can be regarded as the best imaging modality for extracting and understanding macroscopic anatomical information. Since these data may contain a huge and increasing amount of valuable information, it is becoming difficult for radiologists and surgeons to fully utilize the information even with currently available computer assistance. In addition, these data will be potentially useful not only for diagnosis and therapy of the patient to be imaged in the data but also for a wide range of basic and clinical medicine as well as other sciences related to the human body if they are accumulated in the database and processed as population data of human anatomy. Visible Human (VH) data [110] influenced both the medicine and computer science research fields. VOXEL-MAN [5, 111] is a digital (or electronic) anatomical atlas generated from VH data and one of the most successful applications of VH data. Figure 1.12 shows VH data processing (Fig. 1.12a) and typical visualizations of VOXEL-MAN (Fig. 1.12b, c). The resolution of current CT data is high enough to regard anatomical identification

VOXEL-MAN 3D digital atlas reconstructed from Visible Human

Fig. 1.12 VOXEL-MAN 3D digital atlas reconstructed from Visible Human (VH) data. (a) VH data analysis (Ref. [111]). (b) Digital (or electronic) 3D anatomical atlas. (c) 3D reconstruction of the vasculature, bones, chest, abdomen, and pelvis

in CT data as “virtual” dissection of the human body, similar to VOXEL-MAN generated from VH data. Therefore, development of methodologies for automated identification of the whole-body CT data is a key issue for fully utilizing potential information inherent in the whole-body CT data.

Whole-body imaging has influenced radiological diagnosis. In early studies, trade-offs between advantages and disadvantages of whole-body CT screening of healthy individuals have been discussed, and some criticisms have been made because of the radiation dose and limited real benefits [112]. However, in populations with illnesses and a higher pretest likelihood of positive findings, whole-body CT combined with PET is often used for tumor staging [113]. Whole-body PET/CT is now recognized as a useful modality to find unexpected cancers for patients in clinical practice. Apart from cancer diagnosis, whole-body CT has been suggested for serious trauma patients [114]. Whole-body diffusion MRI, which does not involve any radiation, has also been shown to be potentially useful for tumor staging [115] and nerve imaging [116]. More recently, from the technical point of view, dose reduction is becoming possible through innovations in CT reconstruction algorithms [117] and rapid reduction of computational cost in recent years. Therefore, in the

Fully automated segmentation and anatomical identification of whole-body noncontrast CT data. (a) Original CT image. (b) Landmarks. (c) Chest, abdomen, and pelvis. (d) Muscles. (e) Skeletal system

Fig. 1.13 Fully automated segmentation and anatomical identification of whole-body noncontrast CT data. (a) Original CT image. (b) Landmarks. (c) Chest, abdomen, and pelvis. (d) Muscles. (e) Skeletal system

long run, whole-body imaging will become more advantageous, while disadvantages will decrease.

Whole-body imaging is becoming more widespread and opens new research opportunities for the medical image processing field. The concept of multi-organ multi-disease CAD was tested by a Japanese nationwide project [47], which also provided a strong motivation for whole-body image analysis. The importance to computational anatomy is that this involves whole-body 3D data of a large population, rather than of a single subject, such as VOXEL-MAN. Computational anatomy models, which represent intersubject variability of anatomical structures, are systematically constructed from population data, as shown in Fig. 1.13.

The computer vision research field has already developed key technologies for modeling and recognizing anatomical structures from images. In contrast to medical image analysis, which mainly analyzes 3D images inside the body, computer vision deals with outer appearances of persons in addition to other real-world objects. Typical anatomical structures (or parts of the human body) addressed by computer vision include faces, hands, and body motion. Eigenface [118], ASMs [87], and AAMs were typical approaches developed in computer vision. They initially addressed 2D images and then were extended to 3D domains. The morphable face model [119] is one of these extensions and represents intersubject variability of 3D shape of the human face. Similarly, 3D shapes and image intensity distributions of organs and anatomical structures, such as the lung, liver, and pelvis, are modeled by 3D versions of ASMs, SSMs, and AAMs [93]. These technologies will also be essential for whole-body modeling. Although previous approaches assumed single-structure modeling, the body consists of various types of structures, such as the parenchymal organs, vessels, lymphatic system, musculoskeletal system, and nerves, which are interrelated with each other. These structures may be well-described by using different shape modeling schemes. The challenges of whole-body computational anatomy models include the problem of multi-structure modeling of the different types of anatomical structures. It also should be congruent with the discipline of anatomy which has been established for several hundred years. In this book, these efforts will be described.

Education of medical students is one of the important applications of whole- body imaging. As described before, digital or online atlases of the whole body, such as VOXEL-MAN, are now widely available. Furthermore, whole-body autopsy imaging realizes combined virtual and real cadaver dissection. Several papers have evaluated the usefulness of pre-dissection autopsy imaging in the cadaver dissection course in anatomy education [120, 121]. Automated segmentation and anatomical identification of cadaver CT data will enhance the usefulness of pre-dissection autopsy imaging. While conventional digital atlases represent one subject, computational anatomy models typically represent intersubject variability in addition to average shapes. Another potential benefit for education will be to use whole-body parametric computational anatomy modeling, which will generate an individual anatomy having arbitrary height, weight, and variable organ shapes while maintaining consistent relations among anatomical structures. These models will be effective for learning not only typical anatomy but also its variability.

The Physiome is an emerging project for comprehensive modeling of the human body from the aspect of physiology [122]. One of the goals of the Physiome will be patient-specific multi-scale simulations of the whole body to predict patient function after treatment to optimize treatment planning. Whole-body imaging will provide patient-specific anatomy required for multi-scale simulation as aimed at in the Physiome project. In the musculoskeletal system, whole-body analysis is effective in assessing human movement even if the patient has a problem in a specific joint. Patient-specific biomechanical simulations of the whole musculoskeletal system will be meaningful for preoperative prediction of postoperative patient function. Figure 1.14 shows anatomy modeling in OpenSim [123], one of several well- developed platforms for musculoskeletal analysis. As another example, simulation should be performed for the entire cardiovascular and circulatory system [124]. These simulations currently require huge computational power. However, considering rapid reduction of computing cost, the problem will not be permanent. To

OpenSim platform for subject-specific musculoskeletal simulation

Fig. 1.14 OpenSim platform for subject-specific musculoskeletal simulation (Ref. [123]). (a) Anatomy modeling. (b) Simulation realize these scenarios in clinical practice, patient-specific anatomical models need to be reconstructed with sufficient accuracy and a minimum amount of burdens for the medical staff. Therefore, patient-specific modeling from whole-body imaging and other sensory data is regarded as an important issue in the Physiome project.

Whole-body imaging plays an important role in human anatomy modeling for simulations of radiation dosimetry and radio-frequency wave propagation. Several voxel and surface models were developed for this purpose [125-127]. More recently, a whole-body statistical shape model was used to perform simulations of types of the human body [128]. Further improvement of whole-body computational anatomy models will contribute to more accurate, systematic, and comprehensive dosimetry simulations.

Impacts of whole-body imaging on several domains were described, including radiological diagnosis, medical image analysis, simulations for predictive medicine, and dosimetry simulations. Initially, cost and radiation dose were considered to be disadvantageous for clinical practice. However, cost and radiation dose were able to be reduced as a result of technological developments and improvements. Therefore, whole-body imaging will play an increasing role on simulations related to the human body as well as radiological diagnosis. Developments in whole-body computational anatomy and its application to subject-specific anatomy modeling will be a key issue to fully utilize whole-body image data.

< Prev   CONTENTS   Source   Next >