Work

When and why does Computer Aided Detection (CAD) interfere with visual search?

Year: 2011

Type: article

Abstract: The detection of tumors in x-ray images is an important task for radiologists in the battle against cancer. Computer aided detection (CAD) programs are increasingly deployed in the clinic, but these programs are controversial. While most studies show better performance with the help of a CAD system, several studies have suggested that CAD does not increase hits in clinical settings (Gur et al., 2004, Brem & Schoonjans, 2001), while others have shown that CAD increases false alarm rate (Fenton et al., 2007). A common fear is that CAD users may become overly reliant on CAD marks and will miss unmarked cancers. The small amount of data supporting this concern is largely based on behavioral outcomes. In the current study, we monitored eye-position to gain insight into this question. Since radiologists have limited availability as observers, we created a laboratory analog task that could be used with non-experts. Observers searched for semi-transparent Ts amongst Ls. The background was 1/F noise, used as an approximation of visual texture generated by imaging human tissue. Each observer completed 200 trials: one block with the help of a CAD system and one block unassisted. CAD marked 75% of Ts and 10% of Ls. Results: Without CAD, hit rate was 83%. On CAD trials, hit rate was significantly higher if the target was marked (97%, p < .001) but significantly lower if it was not marked (58%, p < .0001). Target fixation rate was comparable for missed targets in CAD and non-CAD conditions. However, dwell time on distractors was lower with CAD. In this task, at least, the increased miss rate for unmarked targets seems to reflect a failure to adequately consider potential targets, rather than a failure to search for and locate them.

Source: Journal of Vision

DOI: 10.1167/11.11.1336

Authors: C. Cunningham, T. Drew, J. M. Wolfe

Institutions: Brigham and Women's Hospital, Harvard University

Language: English

Topic: Radiology practices and education

Subfield: Radiology, Nuclear Medicine and Imaging

Field: Medicine

Domain: Health Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $1,850