From the Viewpoint of Diagnostic Radiology
Before the discovery of X-rays, physicians could not look into live human bodies except during surgery, and the knowledge of anatomy could only be obtained from cadaver dissection and autopsies. X-rays enabled physicians to diagnose and treat diseases based on the findings on roentgenograms using their anatomical knowledge. In interpreting plain radiographs, which are two-dimensional projection images of three-dimensional structures, radiologists apply their knowledge of projected anatomy. For example, radiologists can interpret posterior-anterior mapped view images or lateral mapped view images in chest radiographs. They must then be able to visualize or construct the real anatomical structures from the projected images. This requires considerable training and experience.
Improvements in the technology of radiography, especially oral and intravenous contrast agents, made it possible to image previously radiolucent digestive organs and vessels. The technique of angiography was developed in 1927, 30 years after the discovery of X-rays, by the Portuguese neurologist Egas Moniz, who wished to visualize cerebral circulation . Angiography, which is an imaging technique to visualize blood vessels (mainly arteries but also veins), uses contrast materials injected into vessels to opacify them on X-ray imaging. This allows visualization of arterial stenosis or aneurysm formation. It can visualize many types of arteries, including cerebral arteries and coronary arteries. It can also visualize tumor perfusion by opacifying the feeding artery or arteries. Radiologists can diagnose stenosis of coronary arteries, which causes myocardial infarctions, as well as brain and hepatic tumors. Angiography has developed a range of treatment modalities including balloons and stents to treat stenoses and special catheters to treat tumors with embolization materials.
The development of contrast agents, often combined with air or gas (double contrast) introduced by the oral or rectal routes, enables gastrointestinal radiography examinations such as the upper gastrointestinal (UGI) examination. UGI, which was developed by Shirakabe and Ichikawa, visualizes the esophagus, stomach, duodenum, and sometimes the small bowel, and the barium enema (single or double contrast) visualizes the colon . UGI and barium enema visualize the surface mucosa by opacifying it (single contrast) or coating it (double contrast) with contrast. Radiologists can diagnose stomach and colon diseases with these techniques. In Japan, stomach cancer was the most common neoplasm causing death. However, the number of deaths caused by stomach cancer has decreased, in part because of screening examinations using UGI. New imaging techniques can enable radiologists to diagnose diseases in more organs than was possible with plain X-rays. The interpretation of these images requires a considerable degree of training and experience on the part of the radiologist.
Shinji Takahashi developed what was called “rotational radiography,” which is regarded as an early computed tomographic apparatus in 1953 . Godfrey Hounsfleld furthered the concept of computed tomography (CT) in 1967 and developed the first commercially available CT scanner. The first scan using the commercially available CT scanner was performed in England for brain disease in 1971 . Hounsfleld won the Nobel Prize in medicine in 1979 for the development of CT. CT is a technology that can produce reconstructed images in slice (section) format of the body or brain using a computer to generate Fourier transforms or other reconstruction algorithms. In a CT scanner, a rotating X-ray tube and detector are located on the opposite sides of the target. The detector obtains multidirectional attenuated X-ray data on each rotation and the computer generates an attenuation map image. The quality of early CT images was poor, and scan times were very long. The first brain scan required 4 min to generate only two sections and required 7 min for the calculations to generate a reconstructed image. With the advent of CT, radiologists learned tomographic anatomy. CT images have many advantages over plain radiographs. Most importantly, CT eliminates the superimposition of anatomical structures. CT visualizes differences in X-ray attenuation between soft tissues that would all be radiolucent on plain radiography and constructs images that are attenuation maps. Conventional radiographs can visualize structures such as the skull and lungs in sufficient detail to allow radiologists to diagnose diseases. However, they cannot visualize soft tissues inside the brain or bronchus. Prior to the advent of CT, patients with brain or bronchial lesions had to be examined with pneumoencephalography (using air as a contrast agent) or bronchography (using barium as a contrast agent), which involved invasive procedures involving special training for the radiologist performing them. A large proportion of radiologists’ training is now devoted to the various tomographic modalities.
High-resolution CT (HRCT) images can be obtained using a conventional CT scanner with a high spatial reconstruction algorithm and narrow collimation to maximize resolution. HRCT is especially important in the diagnosis of lung diseases
 . It can reveal minute anatomical structures with high spatial resolution. It can improve the accuracy of diagnosis of diffuse lung diseases and small solitary pulmonary nodules. HRCT can also reveal microanatomic structures such as secondary lobules of lungs. Using HRCT, radiologists can distinguish among different diffuse lung diseases such as interstitial lung diseases and chronic obstructive pulmonary disease, for which microanatomical detail is important for diagnosis.
Spiral (helical) CT (SCT) and multi-detector CT (MDCT) can generate threedimensional volume data. Spiral CT was invented by Kalender in the 1980s
 . HRCT improves spatial resolution in the axial plane. SCT improves spatial resolution along the long axis of the body. Spiral CT, where the X-ray tube rotates continuously as the pallet moves through the gantry at an adjustable pitch, was first used with one row of detectors. The number of rows has increased steadily. MDCT can collect image data for multiple slices at once, decreasing scan time. Current MDCT such as the Aquilion One (Toshiba Medical Systems) has 320 rows of detectors and can scan a 16-cm thickness in only 275 msec. MDCT can image minute structures with reduced motion artifact. The radiologist can obtain anatomical information equivalent to autopsies or anatomical textbooks. Radiologists can improve the accuracy of their diagnoses using a small volume of data, but CT examinations with multiple CT sections have caused many problems for radiologists, resulting in misdiagnoses. To display CT and other digital modalities, most radiologists currently use the DICOM (digital imaging and communications in medicine) standards. DICOM is a standard for handling, storing, printing, and transmitting digital information in medical imaging. Therefore, computer-aided diagnosis (CAD) can be naturally integrated to clinical practice, and many CAD algorithms have been developed. In addition, images from multiple CT sections can help produce digital atlases, which can reveal minute structures in the human body. VOXEL-MAN, which was developed at the University Medical Center Hamburg- Eppendorf, is one of the more famous digital atlases  (Fig. 1.3). Using such digital atlases can help radiologists learn about selected cross sections of the human body in an interactive manner. These atlases can improve the diagnostic abilities of radiologists.
MRI visualizes anatomy and pathology using strong magnetic fields and radiofrequencies to generate images. The advantage of MRI is that it does not involve
Fig. 1.3 VOXEL-MAN 3D-Navigator: Inner Organs. http://www.voxel-man.com/3d-navigator/ inner_organs/
ionizing radiation. MRI is frequently used in the fields of pediatrics, obstetrics, and gynecology. Functional MRI measures signal changes in the brain and can localize brain activity. It is used in neuroscience. Diffusion-weighted MRI (DWI) measures the diffusion of water molecules in tissues and visualizes the connectivity of white matter. These applications are useful for diagnosis of neurological disorders and surgical planning. Real-time MRI can provide radiologists with four-dimensional (spatial and temporal) anatomical data. These data cannot be obtained with CT because of ionizing radiation. These four-dimensional data can provide useful information for diagnosis of conditions such as lung and heart disorders. The abovementioned VOXEL-MAN also includes applications for the creation and visualization of three-dimensional digital models of the human body using crosssectional images from MRI/CT. MRI and CT produce tomographic images, and these are complementarily useful.
Endoscopy is a powerful diagnostic tool for the bronchus, stomach, colon, and other hollow organs. However, endoscopy preparation is often burdensome for patients and requires extensive training for endoscopists. Virtual endoscopy images can be generated using imaging data obtained from SCT or MDCT (mainly MDCT). Virtual endoscopy is different from the abovementioned imaging methods, because it uses post-processing, and synthesizes virtual images. CT colonoscopy is one of the most common virtual endoscopic examinations . CT colonoscopy is less
Fig. 1.4 Detection of abnormality in a breast by an early CAD algorithm (Ref. )
burdensome for patients, and it does not require endoscopy training for physicians (but requires training in interpretation). It is expected to be useful as a screening tool for detection of colorectal polyps. CAD algorithms exist for the detection of polyps in CT colonoscopy . Virtual bronchoscopy is also useful for diagnosis of pulmonary diseases such as lung cancers and has value for surgical preoperative examination .
CAS is one of the most important topics in current image diagnostic diagnosis. CAD provides the quantitative diagnostic information for the detection and classification of tumors for breast cancer, lung cancer, colorectal cancer, and other cancers from medical images such as CT, MRI, and mammograms. The function of CAD is to assist radiologists and improve their diagnostic abilities; its function will not be so-called automatic diagnosis. Early studies of CAD involved mainly mammography and chest radiography [43, 44] (Fig. 1.4), and their goal was detection of abnormal lesions such as those observed in breast cancer and lung cancer. The radiographic image quality and the power of computers were not sufficient for meaningful evaluation of CAD in the early studies [43, 44]. Imaging technology and computer processing power have steadily increased. Various CAD algorithms have also been developed and improved. In 1998, an American CAD company, R2 Technology, was licensed for the first time by the US Food and Drug Administration. This was an important event for CAD development. Radiologists had recognized CAD systems as useful tools in image diagnosis. Subsequently, commercially available CAD systems for detection of colon polyps were produced.
CAD has two roles. One is computer-aided detection (CADe), and the other is computer-aided diagnosis (CADx). Quantitative imaging (QI) is a current trend in medical analysis. To improve the value and practicality of quantitative biomarkers by reducing variability across patients and time, the Quantitative Imaging Biomarkers Alliance (QIBA) was formed by the Radiological Society of North America (RSNA) in 2007. It currently has six active technical committees: MRI, functional MRI, FDG-PET, CT Volumetry, COPD-Asthma, and Ultrasound. Many QI algorithms are now installed on diagnostic workstations in departments of radiology .
CAD systems currently available are for single organs and single diseases, such as detection of calcification and masses on mammography, detection of polyps on CT colonoscopy, and detection and classification of pulmonary nodules on chest X-rays or thoracic CT. These algorithms are mainly based on local image features and machine learning algorithms and are not strictly based on anatomy. This diagnostic logic is different from that of radiologists. Radiologists usually interpret radiographs based on local features and local and global information of anatomical structures. In daily clinical practice, radiologists have to diagnose multiple diseases in multiple organs. For example, in the case of lung cancer, diagnoses are based on primary lesions in the lungs. However, radiologists must also diagnose other pulmonary lesions such as those of diffuse lung disease, and they also need to diagnose metastatic lesions in other organs, such as the liver or brain. In such cases, information regarding multiple anatomical structures is needed. In the Japanese research project entitled “Intelligent Assistance in Diagnosis of Multi-dimensional Medical Images (2003-2007),” researchers developed many CAD algorithms for multiple disease features and entities in multiple organs [46, 47] (Fig. 1.5). At a symposium held in Tokyo in 2007, one of these CAD systems was demonstrated using plug-ins developed by researchers on the common platform named “Pluto.” Many applications of the CAD algorithms were integrated into one system on the common platform as plug-ins.
The computational anatomical models are important for development of local/global CAD systems. Anatomical models such as statistical atlases provide various kinds of information about human organs . Statistical atlases are already popular in the field of neuroscience. One of them is statistical parametric mapping (SPM) . Computational anatomical models are needed for the development of advanced CAD algorithms . The diagnosis made by CAD algorithms
Fig. 1.5 Twelve organs detected by the simultaneous segmentation method (Refs. [46, 47]). Left: Typical axial abdominal CT section with segmented liver and gallbladder. Right: Twelve segmented organs in a 3D display should be based on the computational anatomy as well as the interpretation of the radiologist. These CAD systems, using computational anatomical models, can deal with multiple diseases in multiple organs. Moreover, the high-resolution volume data of CT and MRI can be used to make digital atlases such as VOXEL-MAN. Radiologists can look at selected cross sections of the human body interactively, which they could not do previously. They can improve their knowledge of anatomy using such digital atlases. The computational anatomical models can provide statistical and structural information about human anatomy. Therefore, it is expected that computational anatomy can increase radiologists’ knowledge of anatomy and the functionalities of CAD algorithms.