CHN method region of interest extraction method based on shape information and convolutional neural network

Abstract

The CHN method region of interest extraction method based on the shape information and the convolutional neural network comprises the steps of 1, determining the size of a boundary frame of a reference bone, and 2, positioning key points of the reference bone. The CHN method calculates the bone age by evaluating the maturity indications of 14 reference bones. The method conforms to the growth anddevelopment law of Chinese contemporary teenagers. According to the method, the high universality is ensured while the accurate extraction of the reference bone region of interest is ensured. The reference bone frame size is mainly used for determining the widths of fingers and wrists according to the shape information of the wrist part, and the boundary frame size of 14 reference bones is obtained according to different weights; in the positioning of the key points of the reference bone, regression prediction is carried out mainly through a self-built convolutional neural network model, and key point coordinates of 14 reference bones are obtained. After the frame size and the coordinates of the reference bone are obtained, the regions of interest of the reference bones can be extracted.

Description

technical field [0001] The invention relates to a CHN method interest region extraction method. Background technique [0002] The CHN method calculates bone age by evaluating the maturation indicators of 14 reference bones, which is in line with the growth and development laws of contemporary Chinese adolescents. The 14 reference bones of the CHN method are metacarpal 1, metacarpal 3, metacarpal 5, proximal phalanx 1, proximal phalanx 3, proximal phalanx 5, middle phalanx 3, middle phalanx 5, distal phalanx 1, and distal phalanx 3. Distal phalanx 5, hamate, capitate and radius. [0003] At present, most of the ROI extraction methods are based on edge detection and corner detection. These methods are simple to implement and can accurately extract the ROI of the reference bone, but their placement on the wrist, placement and image quality and other factors are sensitive and have no universal applicability. The present invention not only guarantees the accurate extraction of...

AI Technical Summary

Problems solved by technology

[0003] At present, most of the ROI extraction methods are based on edge detection and corner detection. These methods are simple to implement and can accurately extract the ROI of the reference bone, but their placement on the wrist, placement and image quality Sensitive to factors such as, not universal

Images