Junji Shiraishi, Presenter: Nothing to Disclose

Katsutoshi Sugimoto, Abstract Co-Author: Nothing to Disclose

Fuminori Moriyasu MD, PhD, Abstract Co-Author: Nothing to Disclose

We developed a computer-aided diagnostic (CAD) scheme for detection of hepatic metastases in Sonazoid-enhanced ultrasonography (US).

This study was IRB-approved with written informed consent. Twenty-seven patients with 55 hepatic metastases and 6 patients without hepatic metastasis underwent contrast-enhanced US in the liver-specific phases of Sonazoid. All US scanning was performed with a routine clinical procedure. We used the diagnoses established by contrast-enhanced multi-detector row computed tomography as the standard of reference. We had conducted an observer study for detection of these metastases by 7 radiologists. Among 55 metastases, we excluded 9 metastases for the development of the CAD scheme because the sizes of these lesions were larger than 25mm and could be identified easily by the radiologists in the observer study. In the 1st step of the CAD scheme, sequential view of US scanning within a digital cine clip was separated into single series of scanning by use of a cross-correlation between adjacent video frames. Quasi volume data was produced for each series of scanning. Initial candidates of metastasis were identified by use of the processed volume data in which sphere lesions were enhanced by use of a Hessian matrix filter. Morphologic volume features and grayscale features were extracted for the reduction of false positives (FPs). In addition, we plan to use a multi-layer artificial-neural network (ANN) with extracted image features of candidates for further reduction of FPs.

Our preliminary results indicated that the sensitivities for detection of hepatic metastases were 80.4% (37/46) with 190.2 FPs per patient at the initial pick-up of candidates while average sensitivity obtained by 7 radiologists was 65.5% for these cases. Furthermore, 7 of 12 metastases, which were missed by the majority of 7 radiologists in the observer study, were correctly identified by the CAD. Further reduction of FPs would be expected by the application of the ANN.

The CAD scheme for identifying hepatic metastases by use of contrast-enhanced ultrasonography has a potential to improve diagnostic accuracy.

Clinical utility of contrast-enhanced ultrasonography for the detection of hepatic metastases could be improved by use of the CAD scheme.

Shiraishi, J, Sugimoto, K, Moriyasu, F, Computer-aided Detection of Liver Metastases in Contrast-enhanced Ultrasonography.  Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL. http://archive.rsna.org/2010/9001923.html