Desktop version

Home arrow Health

What Is Computational Anatomy?

Yoshitaka Masutani

Faculty of Information Science, Hiroshima City University, Hiroshima, 731-3194, Japan

e-mail: This email address is being protected from spam bots, you need Javascript enabled to view it

Computational anatomy is an emerging discipline deriving from medical anatomy and several other sciences and technologies, including medical imaging, computer vision, and applied mathematics. The main focus of the discipline covers the quantitative analysis and modeling of variability of biological shapes in human anatomy in health and disease.

Beyond just adding numerical and quantitative information to the conventional anatomy describing human body structures, a wide spectrum of research topics are involved including simulation of average anatomies and normal variations, discovery of structural differences between healthy and diseased populations, detection and classification of pathologies from structural anomalies, and so on.

Disciplines such as morphometrics and anthropology have long been involved with analyzing biological shapes. Among them, the book On Shape and Growth by Thompson published a century ago [1] is regarded as the Bible for morphometrics. It focuses on the importance of the roles of physical laws and mechanics as the fundamental determinants of the form and structure of living organisms.

From the technical viewpoint, statistical analysis of shapes in pattern recognition

[2], computer vision [3], and artificial intelligence [4] can be regarded as one of the origins. In the development of medical imaging research, digitizing data, including spatial and functional relationships of anatomical structures based on high-resolution images of a cadaver [5], was an essential step. It allowed creation of a digital atlas, which has been widely used in medical education. In the medical imaging research field, one of the pioneer works for computational anatomy was initiated by a so-called digital atlas of human anatomy created from high-resolution optical cross-sectional images of a cadaver [5], in which systematically organized knowledge was implemented. The digital atlas is now widely used in medical education.

An important role of computational anatomy in the clinical setting is in computer-assisted diagnosis (CAD) and computer-assisted surgery (CAS). In such application-oriented aspects of the discipline, one of the key demands is computational understanding of medical images with high accuracy and robustness. In other words, reliable and automated segmentation schemes for all organs in medical images are necessary for detecting abnormal structures and surgical planning. It has been a long-term challenge in medical imaging researches and still has been exhaustively studied over time.

Before the 1990s, various automated segmentation techniques based on data- driven approaches using simple techniques such as thresholding and voxel connectivity analysis were developed and were proven to be useful within limited clinical imaging situations.

In the 1990s, shape model-based approaches initiated by SNAKES [6] were intensively used to overcome the drawbacks of data-driven approaches. The concept uses parameter optimization to fit the model to the correct boundary of target organs. Those approaches attained some success. The key to improvement is acquiring statistics of inter-patient variations in representing shapes and image intensities of target organs.

Recent advances in medical image segmentation have mainly been based on using computational models based on a number of organ shape samples. The models are statistical representations of shapes/intensity patterns, called active shape/appearance models (ASM/AAM) [7], which are also known as statistical shape/intensity models. Recently, several mathematical tools, such as the Rieman- nian framework, have been successfully introduced [8], especially for statistical analysis of anatomical structures based on medical images at the population level. Those statistical approaches are effectively combined with machine learning methodologies [9] to obtain more reliable results based on a vast amount of samples.

So far, one of the most successful areas for statistical analysis of image-based anatomy is neuroimaging because of the intensive demand in this field. One of the reasons for this success is related to the challenges involved in investigating brain function. Free software for image analysis such as statistical parameter mapping (SPM) [10] has pushed research in this area forward. In brain image analysis tools, a standard brain atlas, which includes pre-segmented regions, is used. A new dataset is registered to the template and then the pre-segmented regions can be reflected in the data. Such atlas registration techniques can be regarded as another mainstream method in addition to statistical shape/intensity models.

Anatomical structures with nonpathological variations, “anomalies,” can pose problems. For example, the number of the vertebrae is frequently more or less than the normal. That is, there is a discrete variation in the number of the vertebrae. In addition to continuous shape variations, these noncontinuous or discrete organ variations must also be considered in modeling anatomical structures.

Recent achievements based on analysis of a huge number of clinical image samples (so-called big data), including healthy volunteers and real patients, throw new questions at us, such as “What is the definition of abnormality?”, “What is the border between normal and abnormal?”, and “Is a given abnormality significant, or might its workup cause more morbidity without eliminating an important disease?”. The keys to the answers are naturally in the statistical analysis.

In an attempt to answer these questions, a computational anatomy project in Japan was initiated in 2009, supported by a Grant-in-Aid for Scientific Research on Innovative Areas from the Ministry of Education, Culture, Sports, Science, and Technology, Japan [11]. As the project name “Computational Anatomy for Computer-Aided Diagnosis and Therapy: Frontiers of Medical Image Sciences (“Computational Anatomy” (CA) in short)” shows, it was aimed at establishing a mathematical framework to deal with human anatomy and diseases, primarily focused on the chest and abdomen, based on medical images with certain application-oriented aspects such as CAD and surgery.

Several related research projects are found all over the world, such as the Human Connectome Project (HCP) [12] and the Physiome Project [13]. The HCP is intended to construct a human brain map describing the complete neural connections of both structures and functions of intra- and intersubjects (over time). It is a longterm project, begun in 2011, involving the collection and sharing of multimodality images, including magnetic resonance imaging (MRI), among multiple centers. Similarly, the Physiome (“physio-” meaning “life” and the suffix “-ome” meaning “as a whole”) Project is aimed at providing a quantitative description of physiological dynamics and functional behavior of the organism within anatomical structures.

Thus, computational anatomy is a developing research field involving a wide array of sciences and technologies aimed at improving human health.

< Prev   CONTENTS   Source   Next >

Related topics