Computer-aided detection as a decision assistant in chest radiography

Autor: Peter R. Snoeren, Cornelia M. Schaefer-Prokop, Laurens Hogeweg, Bram Platel, Bram van Ginneken, Nico Karssemeijer, Maurice Samulski
Rok vydání: 2011
Předmět:
Zdroj: SPIE Proceedings.
ISSN: 0277-786X
DOI: 10.1117/12.877968
Popis: Background. Contrary to what may be expected, “nding abnormalities in complex images like pulmonarynodules in chest radiographs is not dominated by time-consuming search strategies but by an almost immediateglobal interpretation. This was already known in the nineteen-seventies from experiments with brie”y ”ashedchest radiographs. Later on, experiments with eye-trackers showed that abnormalities attracted the attentionquite fast but often without further reader actions. Prolonging ones search seldom leads to newly found ab-normalities and may even increase the chance of errors. The problem of reading chest radiographs is thereforenot dominated by “nding the abnormalities, but by interpreting them. Hypothesis. This suggests that readerscould bene“t from computer-aided detection (CAD) systems not so much by their ability to prompt potentialabnormalities, but more from their ability to interpret the potential abnormalities. In this paper, this hypothe-sis was investigated by an observer experiment. Experiment. In one condition, the traditional CAD condition ,the most suspicious CAD locations were shown to the subjects, without telling them the levels of suspicious-ness according to CAD. In the other condition, interactive CAD condition , levels of suspiciousness were given,but only when readers requested them at speci“ed locations. These two conditions focus on decreasing searcherrors and decision errors, respectively. Results of reading without CAD were also recorded. Six subjects, allnon-radiologists, read 223 chest radiographs in both conditions. CAD results were obtained from the OnGuard5.0 system developed by Riverain Medical (Miamisburg, Ohio). Results. The observer data were analyzed byLocation Response Operating Characteristic analysis (LROC). It was found that: 1) With the aid of CAD, theperformance is signi“cantly better than without CAD; 2) The performance with interactive CAD is signi“cantlybetter than with traditional CAD at low false positive rates.Keywords: Computer-Aided Detection, Chest radiography, Pulmonary Nodules, Observer Study
Databáze: OpenAIRE