Semi-Supervised Learning in Medical Images Through Graph-Embedded Random Forest

• Published in: Frontiers in neuroinformatics Vol. 14; p. 601829
• Main Authors: Gu, Lin, Zhang, Xiaowei, You, Shaodi, Zhao, Shen, Liu, Zhenzhong, Harada, Tatsuya
• Format: Journal Article
• Language: English
• Published: Lausanne Frontiers Research Foundation 10.11.2020; Frontiers Media S.A
• Subjects: Algorithms; Alzheimer's disease; central nervous system (CNS); Decision trees; Diabetic retinopathy; Entropy; Learning algorithms; Machine learning; manifold learning; Medical imaging; Neuroimaging; Neuroscience; retinal image; semi-supervised learning; vessel segmentation
• ISSN: 1662-5196; 1662-5196
• DOI: 10.3389/fninf.2020.601829

In this paper, we propose a novel semi-supervised random forest to tackle the challenging problem of lacking annotation in medical imaging analysis. Observing that the bottleneck of standard random forest is the biased information gain estimation, we replace it with a novel graph-embedded entropy which incorporates information from both labelled and unlabelled data. Empirical results show that our information gain is more reliable than the one used in the traditional random forest under insufficient labelled data. By slightly modifying the training process of standard random forest, our algorithm significantly improves the performance while preserving the virtue of random forest. Our method has shown superior performance with very limited data in both brain imaging analysis and machine learning benchmark.