Our objective is to create an interactive image segmentation system of the abdominal area for quick volumetric segmentation of the aorta requiring minimal intervention of the human operator. The aforementioned goal is to be achieved by an Active Learning image segmentation system over enhanced image texture features, obtained from the standard Gray Level Co-occurrence Matrix (GLCM) and the Local Binary Patterns (LBP). The process iterates the following steps: first, image segmentation is produced by a Random Forest (RF) classifier trained on a set of image texture features for labeled voxels. The human operator is presented with the most uncertain unlabeled voxels to select some of them for inclusion in the training set, retraining the RF classifier. The approach will be applied to the segmentation of the thrombus in Computed Tomography Angiography (CTA) data of Abdominal Aortic Aneurysm (AAA) patients. A priori knowledge on the expected shape of the target structures is used to filter out undesired detections. On going preliminary experiments on datasets containing diverse number of CT slices (between 216 and 560), each one consisting a real human contrast-enhanced sample of the abdominal area, are underway. The segmentation results obtained with simple image features were promising and highlight the capacity of the used texture features to describe the local variation of the AAA thrombus and thus to provide useful information to the classifier.
J. Maiora, G.A. Papakostas, V.G. Kaburlasos, M. Graña, “A proposal of texture features for interactive CTA segmentation by active learning”, KES International Conference on Innovation in Medicine and Healthcare (InMed-14), San Sebastian, Spain, 9-11 July 2014, pp. 311-320.