Ultrasonic image diagnosis report generation method based on target detection and strategy gradient

Abstract

The invention requests to protect an ultrasonic image diagnosis report generation method based on target detection and strategy gradient, which comprises the following steps: firstly, inputting an image into a target detection model, predicting position information of an organ region, and extracting a feature code of an organ region part according to the predicted position information; inputting the extracted feature codes into a language generation model, decoding the feature codes at different moments and generating words, and finally forming a sentence sequence by the generated words to obtain a final output diagnosis report, wherein the constructed loss function comprises the regional position of the target detection model and the error of the disease information, and the language generation model uses a return function to calculate the negative expected value of the generated diagnosis report and the corresponding label diagnosis report, and the training purpose is to minimize thereturn negative expectation. According to the invention, the diagnosis report corresponding to the ultrasonic image can be generated, and pathological information of the diagnosis report is kept accurate and grammars are natural.

Description

technical field [0001] The invention belongs to the technical field of medical image aided diagnosis, and belongs to a method for generating a medical image diagnosis report. Background technique [0002] Ultrasound images are widely used in the diagnosis and treatment of diseases in clinical practice. The diagnosis report requires high-level doctors to spend a lot of time to write. If the computer algorithm can be used to automatically generate the diagnosis report corresponding to the patient's medical image, it will help to assist Doctor's diagnosis. Ultrasound image diagnostic reports usually focus more on describing the pathological information of abnormal parts, because they can describe the disease category and guide the treatment. If the image area of ​​the abnormal part can be located and analyzed correctly, it will greatly benefit the description of the diagnosis report. There are many different objects in most natural images, and there is visual position informa...

AI Technical Summary

Problems solved by technology

Unlike natural images, ultrasound images are noisy, and the boundaries between the foreground and background of different objects are blurred, making it difficult to accurately identify objects and their location information in ultrasound images

At the same time, medical image diagnosis report algorithms are usually trained using maximum likelihood estimation, and the log-likelihood score of the generated text has little correlation with the quality of the text description. For example, when the log-likelihood score of the generated text is high, the The description quality is not necessarily good

To alleviate this problem, researchers have invented standard grammatical metrics for evaluating text quality, but these metrics are discrete and non-differentiable, and traditional methods are difficult to optimize them

Images