Abstract Archives of the RSNA, 2005
Kenji Suzuki PhD, Presenter: Nothing to Disclose
Hiroyuki Yoshida PhD, Abstract Co-Author: Nothing to Disclose
Janne Johannes Nappi PhD, Abstract Co-Author: Nothing to Disclose
Samuel George Armato PhD, Abstract Co-Author: Nothing to Disclose
Abraham Hillel Dachman MD, Abstract Co-Author: Nothing to Disclose
To develop a method for reduction of false-positive (FP) detections in our computer-aided detection (CAD) of polyps in CT colonography (CTC) by use of multiple massive training artificial neural networks (multi-MTANN).
CTC was performed on 73 patients whose colons were prepared by standard pre-colonoscopy cleansing and were insufflated with room air. Each patient was scanned in both supine and prone positions with either a single- or a multi-detector CT scanner with collimations between 2.5 and 5.0 mm, reconstruction intervals of 1.25-5.0 mm, and tube currents of 60-120 mA. All patients underwent the gold standard optical colonoscopy. Fifteen patients had 28 polyps, 15 of which were 5-9 mm, and 13 were 10-25 mm in size. The CTC cases were then subjected to our previously reported CAD scheme that included centerline-based extraction of the colon, shape-based detection of polyps, and reduction of FPs by use of a Bayesian neural network based on geometric and texture features. To eliminate FPs further, we developed a multi-MTANN consisting of four 3-dimensional (3D) MTANNs that were designed to reduce four representative causes of FPs: folds, stool, ileocecal valves, and rectal tubes. To differentiate polyps from FPs (non-polyps), the multi-MTANN was trained with the original CTC volumes as input, and a 3D Gaussian distribution and zeros as teaching volumes for polyps and non-polyps, respectively. Each MTANN was trained with 10 representative polyps and 10 non-polyps in each of the four FP types which were obtained from an independent data set. Outputs of the four MTANNs were combined by the logical multiplication operation for obtaining the final detection result.
The original CAD scheme yielded 96.4% (27/28) by-polyp and 100% by-patient detection sensitivities, both with 3.1 (224/73) FPs per case. The multi-MTANN removed 50.4% (113/224) of FPs without loss of any true positives. As a result, the FP rate of our CAD scheme was improved to 1.5 (111/73) FPs per case while the original sensitivity was maintained.
Application of the multi-MTANN can significantly reduce the number of FPs in the computerized detection of polyps in CTC while maintaining a high sensitivity.
A.H.D.: Consultant, Research support, E-Z-EM; Consultant, GE Healthcare.H.Y.,S.G.A.: Shareholders, R2 Technology, Inc.
Suzuki, K,
Yoshida, H,
Nappi, J,
Armato, S,
Dachman, A,
False-positive Reduction in Computer-aided Detection of Polyps in CT Colonography Based on Multiple Massive Training Artificial Neural Networks. Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL.
http://archive.rsna.org/2005/4407623.html