Desktop version

Home arrow Health

Computer-Assisted Diagnosis

CAD has been rapidly developing over the past three decades. Using dedicated computer systems, CAD interprets medical images and provides a “second opinion.” The final medical decision, however, is ultimately made by a physician [1-3]. Studies on CAD systems and technology reveal that CAD improves diagnostic accuracy of physicians and lightens the burden of increasing workload. Moreover, CAD reduces the number of lesions overlooked because of fatigue or high volumes of data and improves inter- and intrareader variability. CAD has been successfully applied in a variety of clinical areas such as mammography, chest radiography and CT, and CT colonography. All of these technologies are approved by the US Food and Drug Administration (FDA) and are commercially available in the USA and some other countries. These CAD systems are classified as computer-aided detection (CADe) systems and can be differentiated from so-called computer-aided diagnosis (CADx) systems, which evaluate conspicuous structures and determine whether they are benign or malignant.

Enlisting the assistance of a computer to analyze medical images is not a new idea. In fact, in 1963, Lodwick et al. investigated using a computer to diagnose bone tumors [4]. In 1964, Meyers et al. proposed a system to automatically distinguish normal chest radiographs from abnormal ones by measuring the cardiothoracic ratio

[5]. In 1967, Winsberg et al. employed optical scanning to detect radiographic abnormalities in mammograms [6]. Although these earlier systems differ from CAD as we know it today, most of these researchers were well on their way to designing automated diagnostic systems. One of the most important years in the history of CAD was 1998. This year marked the transition of CAD technologies from the research phase to clinical practice with the success of ImageChecker (R2 Technology, Inc., Sunnyvale CA, USA; later acquired by Hologic, Inc., Bedford MA, USA in 2006), which obtained FDA approval. ImageChecker is a computer system intended to mark regions of interest (ROIs) on routine screening mammograms. Today, it is estimated that more than 10,000 mammography CAD systems are used by hospitals, clinics, and screening centers across the USA.

CAD systems target lesions in every human organ and tissue and work with every imaging modality currently in clinical use. Many of the CAD technologies developed to date rely on brute force. For CADe, this means systematically locating regions displaying lesion-like features. With breast mass detection, for example, CADe systems search for regions having the potential characteristics of a tumor by looking for oval shapes with a certain level of contrast. Similarly, CADx systems measure and analyze malignancy characteristics of candidate regions (e.g., the state of the edges) using feature values. Incorporating the new concept of CA into existing CAD systems is very rare [7]. Regardless, new techniques are being developed and applied, to a limited extent, to basic research in image analysis and as a component of CAD systems.

While a number of CAD systems have been implemented for clinical use, several technical problems exist that must be addressed. In CADe systems, false-positive (FP) cases, where the computer erroneously detects a lesion, always exist. The FP rates in current CADe systems are known to be five to ten times greater than those of physicians. To further reduce the FP rate and enhance CADe performance, new techniques must be incorporated that consider anatomical structural information in regions surrounding candidate lesions. For example, if a CADe system can accurately determine whether a feature found in a thoracic CT is a nodule or part of a vessel, the FP rate can potentially be greatly reduced. For this to be possible, a technique based on anatomical information that accurately differentiates between the two types of vessels (arteries and veins) is required.

A more sophisticated CAD system would be one with the capability of identifying lesions across multiple organs. The system would first need to determine where the target organ or organs are located within the image. In the case of multiple organs, information regarding their location in relation to each other would also be required. Techniques that can address issues such as automatic landmark detection, a probabilistic atlas construction method, a statistical shape model (SSM) construction method, and utilization of an anatomical knowledge database are explained in Chap. 2. These techniques must be expressed mathematically to achieve these goals.

It is expected that techniques based on CA will lead to the development of next- generation CAD systems that are even more sophisticated, with the capability to handle multiple organs and multiple diseases.

< Prev   CONTENTS   Source   Next >

Related topics