Ayako Yagahara, Presenter: Nothing to Disclose

Naoki Nishimoto MS, Abstract Co-Author: Nothing to Disclose

Kensuke Fujiwara, Abstract Co-Author: Nothing to Disclose

Tsuyoshi Kitagawa, Abstract Co-Author: Nothing to Disclose

Gou Inoue, Abstract Co-Author: Nothing to Disclose

Katsuhiko Ogasawara PhD, Abstract Co-Author: Nothing to Disclose

At RSNA 2009 we proposed that Improved Dual Snakes (IDS) achieved improved precision compared to that of conventional contour extraction algorithm, “snakes”, and reported that the precision of outline extraction was better than snakes against the phantom images for brain CT. IDS uses two initial contours, one inside and the other outside. The outer contour is deformed using traditional snakes energy and then the inner one is spread using area and spring energy. However, problems remain that the parameter of IDS depends on the shape and the density of objects. The purpose of this study is to apply IDS to brain MRI images. We categorized parameters of IDS using feature quantities of the temporal horns in the lateral ventricles.

IDS was applied 10 MRI images (3 male, 20.8±2.2 y/o). To assess the variation between IDS and the Gold Standard, which was obtained by manual extraction of four radiological technologists, we used precision, recall and F-measure. In each image, parameters of the highest F-measure were set manually. Clustering was performed using feature values extracted from ImageJ and parameters were determined for each cluster. The feature values are as follows: area, mean gray value, SD, modal gray value, min and max gray value, median, perimeter, width and height of bounding rectangle, major and minor axis of oval figure, circularity, Feret's Diameter, skewness, and kurtosis.

The average values of F-measures were 0.79±0.05. Manual extraction of the average of F-measure was 0.76 ± 0.10. Consequently, F-measure of IDS is considered to agree with that of manual extraction. We created three clusters. Tendency of the feature values of each cluster are as follows: one has a large area with long width of bounding rectangle, another has great modal gray value with low kurtosis, and the other has great max gray value, small area, small modal gray value, and low kurtosis. Created from the clusters, the combinations of parameters were obtained with 5 patterns and the results of applying MRI images to IDS, the average of F-measure was 0.79±0.03.

The accuracy of IDS is equivalent to that of manual extraction. We were able to classify the five combinations of parameters using clustering.

This study will contribute to the domain of computer aided diagnosis. This method is a useful key to an early diagnosis of morphological changes of the hippocampus.

Yagahara, A, Nishimoto, N, Fujiwara, K, Kitagawa, T, Inoue, G, Ogasawara, K, Automatic Extraction of the Temporal Horns of the Lateral Ventricles Using Improved Dual Snakes on MRI Images and Classification of Parameters.  Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL. http://archive.rsna.org/2010/9004779.html