We evaluated the performance of an automated polyp-detection scheme on both low-dose and normal-dose virtual colonoscopy data from different institutions. The polyp-detection algorithm is statistically trained on low-dose virtual colonoscopy data from 13 patients with a predicted sensitivity of 92% and a false-positive rate of 9 objects per study. An independent test on data from 50 patients with similar preparation and acquisition as the training data and 32 patients having completely different preparation and acquisition confirms the prediction. Our results show that it is possible to design a sensitive and specific polyp-detector that is robust for low-dose and normal-dose, isotropic and anisotropic, for faecal-tagged and non-tagged CT colonoscopy data.
|Name||International Congress Series|
|Conference||CARS 2005 : International Congress and Exhibition Computer Assisted Radiology and Surgery|
|Abbreviated title||CARS 2005|
|Period||22/06/05 → 25/06/05|
|Other||CARS 2005 : International Congress and Exhibition Computer Assisted Radiology and Surgery ; 19 (Berlin) : 2005.06.22-25|