Sarah Poetter-Lang, Abstract Co-Author: Nothing to Disclose

Diederick Willem de Boo MD, Abstract Co-Author: Nothing to Disclose

Cornelia Maria Schaefer-Prokop MD, Abstract Co-Author: Nothing to Disclose

Soeren Harald Kreuzer MD, Abstract Co-Author: Nothing to Disclose

Michael Weber PhD, Abstract Co-Author: Nothing to Disclose

Martin Uffmann MD, Presenter: Nothing to Disclose

To evaluate the performance of a computer aided detection system (CAD) as a second reader, depending on the readers’ experience and the visibility of artificial lesions.

The study group comprised 116 pa CXR (a-CSI-TFT). Fifty-seven of them contained one or multiple superimposed lesions, simulating nodules of different size (5-15mm), density and contour sharpness. Images were evaluated without and with CAD (xLNA Enterprise 2.0, EDDA systems, Philips) used as a second reader. Four readers of vastly different experience with respect to reading CXRs (A and B: <1 yr, C: >5 yrs, D: > 15 yrs) and experience with CAD (A and B: < 6 wks, C and D: >12 months) were asked to assess the presence of nodules and to localize them. The rating scores were tested for statistical differences using analysis of variance with repeated measures.

Lesion detection was dependent of anatomical location, size and shape (p<0,01). The sensitivity of the experienced readers was significantly higher compared to the less experienced readers irrespective of the use of CAD (with CAD 60,7% vs. 50,8%, without CAD 59,4% vs. 49,9; p<0,001). For a subset of borderline visible lesions the sensitivity significantly increased with CAD for both the experienced readers and for the less experienced readers (with CAD 42,9% vs. 28,8%, without CAD 38,3% vs. 27,0%; p<0,05). The number of false positive lesions increased for all readers with CAD (p<0.05).

We conclude that CAD does increase the detection of subtle lesions on CXRs independent from the reader´s experience. CAD is likely to decrease readers’ specificity.

Observer studies can demonstrate that CAD is helpful in the detection of borderline visible lung lesions in both experienced and less experienced readers.

Poetter-Lang, S, de Boo, D, Schaefer-Prokop, C, Kreuzer, S, Weber, M, Uffmann, M, Performance of CAD for the Detection of Artificial Lung Nodules in CXR.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8016487.html