Visual Search in Medical Images
Harold L Kundel
DOI: 10.1117/3.832716.ch18
text A A A


18.1 Introduction

18.1.1 Definition of search tasks

Searching is a common human activity. We search for a book in the library by looking systematically along the shelves; we search for a friend in a crowd by looking for a familiar face; we search for a pin on the floor by crawling around and feeling for the pin. The common elements in search tasks are a known object that is the purpose of the search, some knowledge about how and where the object is hidden, and a deliberate concentration on the performance of search. In technical terms, search involves a target, prior-knowledge about the target-background relationship, and selective attention. Visual search implies that the visual perceptual system is being used to conduct the search.

18.1.2 Searching for targets in medical images

This chapter will focus on searching for small targets in medical images. Tumors are commonly used as targets in studies of visual search because they rank highly among the abnormalities that are overlooked by radiologists in everyday practice [1]. An investigator who wishes to study tumor detection can find many examples of tumors that are small and inconspicuous. (The terms “nodule” and “mass” are frequently used when referring to tumors and will be used interchangeably in this chapter without any specific nuance of meaning implied. They are all solid lumps of tissue. Nodule is frequently used for lung and liver tumors whereas mass is used for to breast tumors.) However, images with naturally occurring tumors may not be entirely suitable for use as stimuli because the tumors may either occur in combination with other distracting abnormalities or may vary too much in physical characteristics. To overcome these disadvantages, tumors are frequently simulated on carefully chosen background images. Simulation makes available large numbers of tumors with well defined physical characteristics that can be placed anywhere on selected backgrounds. Pulmonary nodules have been simulated on chest images optically [2, 3] and digitally [4, 5]. Microcalcifications and masses have been simulated on mammograms [6] and tumors have been simulated in computed tomograms of the liver [7]. Some critics have argued that simulated lesions are not sufficiently authentic because subtle characteristics might not be included [8]. As a response to such criticism and using the power of computer-illustration programs, investigators have cut lesions out of abnormal images and transplanted them into normal images to produce “hybrids” that come closer to reality than directly synthesized lesions [6].

18.1.3 Imaging tasks

Search is an important part of many of the more general tasks carried out by diagnostic radiologists. Diagnostic imaging tasks can be divided roughly into the five categories shown in Table 18.1: detection, comparison, location, classification, and estimation. Detection and location certainly involve search, and the other tasks might involve search as well.

© 2000 Society of Photo-Optical Instrumentation Engineers

Access This Chapter
Please Wait... Processing your request... Please Wait.
Sign In

Related Content

Customize your page view by dragging & repositioning the boxes below.

Related Book Chapters

Topic Collections


Access to SPIE eBooks is limited to subscribing institutions and is not available as part of a personal subscription. Print or electronic versions of individual SPIE books may be purchased via SPIE.org.

Your library does not currently subscribe to eBooks on the SPIE Digital Library. Print or electronic versions of individual SPIE books may be purchased via SPIE.org.

Sign In