Medical image recognition method based on Knet deep learning

Abstract

The invention provides a medical image recognition method based on Knet deep learning. The medical image recognition method comprises the following steps: A, automatically cutting an effective area from an ultrasonic image; B, recognizing the automatically cut ultrasonic image by adopting Knet: carrying out convolution feature extraction processing on the ultrasonic image; C, carrying out deconvolution feature superposition processing, so that the position of the target feature in the output prediction result M is enhanced to form an enhanced area; D, gradually and independently extracting each piece of convolution feature extraction data Z in the encoder, and carrying out shape specification unification processing on an output prediction result M; E, obtaining a gaze area mapping result M'with the same shape deformation, multiplying each piece of convolution feature extraction data Z by the gaze area mapping result M 'to obtain a gaze feature graph F in a superposed form, namely, displaying a strengthened area and a corresponding area of an input initial ultrasonic image in a superposed manner, and performing same processing by a decoder; F, performing subsequent upgrading on a K_net system.

Description

technical field [0001] The invention belongs to the technical field of computer artificial intelligence, and relates to an algorithm of neural network deep learning, in particular to a medical image recognition method based on K_net deep learning. Background technique [0002] Deep learning is a new research direction in the field of machine learning, which is introduced into machine learning to make it closer to the original goal - artificial intelligence. The performance of deep learning of computer neural networks has developed significantly in recent years. From face recognition to automatic operation, it is possible to provide convenience in all aspects of our lives. [0003] In daily medical work and clinical trials, the application of medical imaging is very popular, and the application method of artificial neural network has also provided great help in medical imaging. Ultrasound images and ranging images, which are often used in initial diagnosis, are relatively ea...

AI Technical Summary

Problems solved by technology

However, the existing U-Net technology still has problems such as inaccurate segmentation of abnormal images, insufficient accuracy of pathological judgments, and low operating efficiency of deep learning.

[0006] In addition, the existing B-ultrasound images are all fan-shaped structures, and the black negative film on the periphery of the image is an invalid graphic area. In the prior art, in the automatic recognition of the computer, the black negative film on the periphery is also subjected to a conventional recognition operation, thus occupying the computer. Memory and running time, reduce work efficiency

Images