Dorith Shaham MD, Presenter: Research Consultant, Siemens AG

Naama R. Bogot MD, Abstract Co-Author: Research Consultant, Siemens AG

Sivan Lieberman MD, Abstract Co-Author: Nothing to Disclose

Ruth Eliahou MD, Abstract Co-Author: Nothing to Disclose

Shaliendra S. Mansur, Abstract Co-Author: Research Consultant, Siemens AG

Alexandra Manevitch, Abstract Co-Author: Employee, Siemens AG

Issac Leichter, Abstract Co-Author: Employee, Siemens AG

M.S. Dinesh, Abstract Co-Author: Employee, Siemens AG

Mausumi Acharyya PhD, Abstract Co-Author: Nothing to Disclose

Jonathan Stoeckel PhD, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To evaluate whether a CAD device improves the detection rate of lung nodules on digital chest radiographs by interpreters with various levels of expertise.

Seventy seven digital chest radiographs (DR) recruited from 3 sites were analyzed retrospectively by two independent expert readers, who correlated DR to chest CT to establish the Ground Truth for the presence and location of masses and lung nodules >5 mm. Disagreements were resolved by concensus. The cases were read in a different random order by 3 panel radiologists with various levels of expertise (resident, general radiologist and thoracic radiologist), who independently analyzed the DR images and marked apparent lung masses and nodules >5mm without CT correlation. A prototype detection algorithm (Siemens) which generated CAD marks was applied to all DR images. Assuming that all CAD marks are accepted by the readers as true marks, we evaluated whether the CAD device, used as a second reader, would improve the readings of each of the panel radiologists.

A total of 46 findings in 36 digital radiographs were identified as the Ground Truth, 39 of which were nodules (5-30mm), found in 29 radiographs. The CAD device improved the detection rate of nodules but not of masses and generated 1.7 false marks per radiograph. The resident's detection rate for nodules improved with the prototype CAD device from 14 nodules in 29 radiographs (48.3%) to 18/29 (62.1%). For the general radiologist, detection improved from 20/29 (69.0%) to 23/29 (79.3%) and for the thoracic radiologist from 20/29 (69.0%) to 22/29 (75.9%). Thus the CAD device improved the detection rate of the resident by 13.8% (p<0.046), of the general radiologist by 10.3% (p<0.083) and of the thoracic radiologist by 6.9%.(p<0.157).

The CAD device increased detection rate for all readers but increase was significant only for the resident. The increase was inversely proportional to the expertise of the reader.

A CAD device for detecting lung nodules on digital chest radiographs improves sensitivity, particularly of less experienced readers, and can be used as a second reader.

Shaham, D, Bogot, N, Lieberman, S, Eliahou, R, Mansur, S, Manevitch, A, Leichter, I, Dinesh, M, Acharyya, M, Stoeckel, J, et al, , et al, , Does CAD Improve the Detection of Lung Nodules on Digital Chest Radiographs by Readers with Various Levels of Expertise?.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5006930.html