Progressive bone age assessment method based on multi-granularity feature fusion

Abstract

The invention relates to a progressive bone age assessment method based on multi-granularity feature fusion. The granularity grading module based on a random puzzle mode is constructed, granularity information contained in an input picture is graded from fine to coarse, and rich local features of all parts of a hand bone are learned through a network; and by constructing a progressive multi-scale feature fusion module, the network is iterated for multiple times, so that global features and local features with most differentiated positions can be learned, other local features can be fused, features containing different granularity information can be learned finally, and the performance and generalization ability of a bone age evaluation model are greatly improved. According to the method, the most distinguished RoIs local features can be concerned, other local features with different granularities can be fused together in a collaborative mode, and higher robustness is achieved.

Description

technical field [0001] The invention relates to the field of medical image processing, in particular to a progressive bone age assessment method based on multi-granularity feature fusion. Background technique [0002] Bone age is an important indicator to reflect the growth and development of adolescents, and is an important content to determine the biological age of the human body. The traditional bone age assessment method is to manually evaluate the hand bone X-ray film of the subject according to the evaluation standards such as G-P, TW-3 or Zhonghua-05. The process is relatively cumbersome and is less affected by individual subjective factors. Large, and it is difficult to learn from the experience among doctors, which requires higher skills for relevant practitioners. [0003] Using the computer-aided bone age assessment method of machine learning, by extracting the shape and texture features of the position with high feature contribution in the hand bone X-ray image ...

AI Technical Summary

Problems solved by technology

However, in the process of continuous growth of the hand bones, each part is changing, large or small, they all contain a lot of prior knowledge, and the relationship is close, if only the most distinguishing RoIs are considered, then the local parts of other RoIs Features will be ignored, which will cause the model to fall into an overfitting situation, resulting in poor generalization ability

Images