Full convolution genetic neural network optimization method for infant brain medical record image segmentation

Abstract

The invention discloses a full convolution genetic neural network method for infant brain medical record image segmentation, and belongs to the field of medical image information intelligent processing, and the method comprises the steps: firstly inputting infant brain medical record image data, carrying out the preprocessing of an image, and carrying out the genetic coding initialization of a parameter according to the weight length L of a DMPGA-FCN network; randomly dividing m individuals into a genetic primitive sub-population Pop and deriving a twin sub-population Pop', determining respective exchange probabilities pc and mutation probabilities pm of the sub-populations in a disjoint interval, and searching an optimal initial weight fa by using a genetic operator; secondly, wherein fais used as a forward propagation calculation parameter, and weighted Q operation is carried out on a feature address featuremap; and finally, performing pixel-by-pixel cross entropy loss calculation on the infant brain medical record prediction image and the standard segmentation image so as to reversely update the weight, and finally obtaining the optimal weight of the infant brain medical recordimage segmentation network model. The method can improve the infant brain medical record image segmentation efficiency, and is of great significance to the early correct diagnosis of infant encephalopathy and the rehabilitation of child encephalopathy.

Description

technical field [0001] The invention relates to the field of intelligent processing of medical image information, in particular to a fully convolutional genetic neural network method for image segmentation of infant brain medical records. Background technique [0002] Infancy is a critical period for brain development. It not only develops rapidly, but also has strong plasticity. The probability of suffering from various encephalopathy is also much higher than that of adults or other growth periods. In recent years, the survival rate of high-risk infants and very low birth weight infants has been greatly improved, and the incidence of brain diseases in infants and young children has shown an upward trend, making early diagnosis more difficult. Therefore, it is of great significance for the diagnosis and evaluation of infant brain diseases to actively explore computer intelligence-assisted early diagnosis methods for infant brain diseases. [0003] In recent years, with the ...

AI Technical Summary

Problems solved by technology

However, due to the characteristics of low resolution and non-uniform gray scale of images of infantile brain medical record images, when using traditional full convolutional neural network methods to segment infantile brain medical record images, gradient descent is prone to occur, and the algorithm is difficult. Significant problems such as local optimal solution and inability to effectively retain image feature information lead to these methods in the segmentation of infant brain medical records, such as time-consuming, difficult training, and low accuracy, so that they cannot obtain highly accurate infant brain images. Medical record segmentation image

Images