Micro-structural tissue analysis for automatic histopathological image annotation

作者：Diaz Gloria; Romero Eduardo^*

来源：Microscopy Research and Technique, 2012, 75(3): 343-358.

DOI：10.1002/jemt.21063

• semantic annotation
• histopathology
• part-based representation
• latent semantic analysis
• local descriptors

This article presents a new approach for extracting high level semantic concepts from digital histopathological images. This strategy provides not only annotation of several biological concepts, but also a coarse location of these concepts. The proposed approach is composed of five main steps: (1) a stain decomposition stage, which separates the contribution of hematoxylin and eosin dyes, (2) a color standardization that corrects color image differences, (3) a part-based representation, which describes the image in terms of the conditional probability of relevant local patches, selected by their stain contributions, (4) a discriminative classification model, which bridges out the found patterns and the biological concepts, (5) a block-based annotation strategy that identifies the multiple biological concepts within an image. A set of 655 skin images, containing 10 biological concepts of skin tissues were used for assessing the proposed approach, obtaining a sensitivity of 84% and a specificity of 67% when annotating images with multiple concepts. Microsc. Res. Tech., 2011.

• 出版日期2012-3