Authors: Samira Bagheri; Hedieh Sajedi
Addresses: Department of Computer Science, School of Mathematics, Statistics and Computer Science, College of Science, University of Tehran, Iran ' Department of Computer Science, School of Mathematics, Statistics and Computer Science, College of Science, University of Tehran, Iran
Abstract: Image annotation is one of the aspects of automatic classification algorithms. In this paper, we propose a learning-based algorithm for automatic medical image annotation based on local appearance cues by using integral images, Gabor transforms, AdaBoost and support vector machine (SVM) classification algorithm. The algorithm starts with a number of landmarks. By landmarks, we generate special sub-patches and extract features using Gabor transforms. In fact, Gabor filter is applied to integral image. AdaBoost in most cases can make a decision about the instances at this stage. To solve the misclassified cases, we combine similar classes and employ SVM/AdaBoost to recognise close instances. This approach is evaluated on large-scale radiograph images and several experiments have been done. At first, we specify five classes with the most number of images in ImageCLEF2008. Then, the proposed method is used for position identification task. At the end, our approach achieved the acceptable accuracies for multiclass radiograph annotation task, when compared to other algorithms. We obtain the accuracy of 95% for multiclass radiograph annotation and >99% for position identification task.
Keywords: medical image annotation; landmark; integral image; Gabor transform; AdaBoost algorithm; support vector machine.
International Journal of Intelligent Machines and Robotics, 2019 Vol.1 No.3, pp.237 - 252
Received: 19 Sep 2018
Accepted: 05 Jan 2019
Published online: 26 Aug 2019 *