Medical image automatic annotation method under small sample condition

Abstract

The invention relates to a medical image automatic annotation method under a small sample condition, and the method comprises the following steps: building a dense linear regression model and a sparse linear regression model for shape mark points of a small sample medical image tissue organ, and synthesizing shape mark points of sample image tissue organs corresponding to other positions; building an affine transformation model for a shape mark area of the small sample medical image tissue organ, and synthesizing textures of the sample image tissue organs corresponding to the other positions; performing a principle component analysis for tissue organs of a new synthesized sample and an original sample image, and generating shape and texture deformable models for the tissue organs; automatically annotating image content on a to-be-annotated image according to astringency of the parameters such as translation, rotation and pantograph ratio of the shape and texture deformable models for the tissue organs by using initial location of a Hough vote study method.

Description

technical field [0001] The invention relates to a method for computer processing of medical images, in particular to a method for automatic labeling of medical images under the condition of small samples, which is suitable for computer medical image synthesis, image retrieval and image labeling. Background technique [0002] Medical images are the most important auxiliary means for medical diagnosis, drug reaction monitoring, and disease management. They have the advantages of fast speed, non-invasiveness, small side effects, low cost, and good effects. As the number of medical images grows exponentially, it becomes increasingly difficult for ordinary users to retrieve the images they need. In the past few decades, there have been a large number of researches on medical image retrieval, which can be summarized into three categories: text annotation methods, content-based image retrieval methods and automatic image annotation (Automatic Image Annotation, AIA) methods. Giild'...

AI Technical Summary

Problems solved by technology

[0004] (1) Many existing methods require a large number of training samples, and manually mark points for training images, which is time-consuming and error-prone, and loses the value of clinical application

[0005] (2) Due to the increase of noise data and the imbalance of the number of training samples between classes, it is easy to reduce the quality of medical image annotation

Images