Ultrasonic image automatic description method and system

Abstract

The invention discloses an ultrasonic image automatic understanding and generation method, and relates to the technical field of computer information processing. The method comprises the steps of extracting an ultrasonic image eigenvector by a CNN model, performing word segmentation on a descriptive sentence corresponding to an ultrasonic image by a word segmentation tool, and obtaining a word sequence eigenvector; training an LSTM model by utilizing the ultrasonic image eigenvector and the corresponding word sequence eigenvector; and generating a plurality of proper sentences at a time step by the LSTM language model through utilizing a beam search method, and generating the sentence for describing the ultrasonic image according to probability estimation. The method can well solve the problem that adjustment cannot be carried out in time when a new combination situation occurs in conventional image topic description and automatic generation. The method can be widely used for understanding and statement description of the ultrasonic image.

Description

technical field [0001] The method relates to the technical field of pattern recognition, specifically a method for automatically describing ultrasonic images in combination with a convolutional neural network model (CNN) and a long-term short-term memory model (LSTM) in a deep learning algorithm. Background technique [0002] The ultrasonic image automatic understanding generation method refers to automatically generating sentences describing the content of the ultrasonic image according to the input ultrasonic image. Nowadays, the method of automatically generating sentences describing the content of ultrasound images based on the content of ultrasound images is gradually being paid attention to by more and more scholars, but there are still some technical difficulties: 1) Due to the complexity of the ultrasound image itself, the traditional feature Extraction methods, such as: SIFT, LBP, SURF, etc., cannot capture important image information in ultrasound images well; 2) T...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the problems existing in the automatic generation of ultrasonic image descriptions, such as the difficulty of capturing ultrasonic image features, the capture of lesion information in ultrasonic images, and the combination of new situations in ultrasonic images, the present invention uses the ultrasonic image and the text content describing the content of the ultrasonic image directly as The training sample pair is used to train the LSTM model, which improves the ability to capture the characteristics of ultrasound images, and can also solve the problems of the traditional automatic generation of description image content, such as the emergence of new combinations of diseases

Images