Efficient Learning by Combining Confidence-Rated Classifiers to Incorporate Unlabeled Medical Data

Abstract. In this paper, we propose a new dynamic learning framework that requires a small amount of labeled data in the beginning, then incrementally discovers informative unlabeled data to be hand-labeled and incorporates them into the training set to improve learning performance. This approach has great potential to reduce the training expense in many medical image analysis applications. The main contributions lie in a new strategy to combine confidence-rated classifiers learned on different feature sets and a robust way to evaluate the "informativeness" of each unlabeled example. Our framework is applied to the problem of classifying microscopic cell images. The experimental results show that 1) our strategy is more effective than simply multiplying the predicted probabilities, 2) the error rate of high-confidence predictions is much lower than the average error rate, and 3) hand-labeling informative examples with low-confidence predictions improves performance efficientl...

• Type: Conference
• Year: 2005
• Where: MICCAI
• Authors: Weijun He, Xiaolei Huang, Dimitris N. Metaxas, Xiaoyou Ying
• Keywords: Informative Unlabeled Data, Medical Imaging, MICCAI 2005, Predictions Improves Performance, Unlabeled Example