 |
Journal articles
Pulmonary nodule detection on MDCT images: evaluation of diagnostic performance using thin axial images, maximum intensity projections, and computer-assisted detection.
Abstract : This study aimed at evaluating the diagnostic benefits of maximum intensity projections (MIP) and a commercially available computed-assisted detection system (CAD) for the detection of pulmonary nodules on MDCT as compared with standard 1-mm images on lung cancer screening material. Thirty subjects were randomly selected from our database. Three radiologists independently reviewed three types of images: axial 1-mm images, axial MIP slabs, and CAD system detections. Two independent experienced chest radiologists decided which were true-positive nodules. Two hundred eighty-five nodules > or =1 mm were identified as true-positive by consensus of two independent chest radiologists. The detection rates of the three independent observers with 1-mm axial images were 22 +/- 4.8%, 30 +/- 5.3%, and 47 +/- 2.8%; with MIP: 33 +/- 5.4%, 39 +/- 5.7%, and 45 +/- 5.8%; and with CAD: 35 +/- 5.6%, 36 +/- 5.6%, and 36 +/- 5.6%. There was a reading technique effect on the observers' sensitivity for nodule detection: sensitivities with MIP were higher than with 1-mm images or CAD for all nodules (F-values = 0.046). For nodules > or =3 mm, readers' sensitivities were higher with 1-mm images or MIP than with CAD (p < 0.0001). CAD was the most and MIP the less time-consuming technique (p < 0.0001). MIP and CAD reduced the number of overlooked small nodules. As MIP is more sensitive and less time consuming than the CAD we used, we recommend viewing MIP and 1-mm images for the detection of pulmonary nodules.
https://www.hal.inserm.fr/inserm-00348861
Contributor : Aurélien Vesin <>
Submitted on : Monday, December 22, 2008 - 2:55:09 PM
Last modification on : Thursday, August 27, 2020 - 11:36:03 AM
Contributor : Aurélien Vesin <>
Submitted on : Monday, December 22, 2008 - 2:55:09 PM
Last modification on : Thursday, August 27, 2020 - 11:36:03 AM
Links full text
