Image report structured extraction method
An extraction method and structured technology, applied in the field of structured extraction of image reports, can solve the problems of low recognition accuracy and slow model convergence speed, reduce the impact of input errors, ensure information accuracy, and improve word segmentation accuracy. Effect
Active Publication Date: 2022-04-12
浙江卡易智慧医疗科技有限公司
View PDF11 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] The invention mainly solves the problems of slow convergence speed and low recognition accuracy of the image report entity text extraction in the existing technology; provides a structured image report extraction method, speeds up the convergence speed of the model and improves the accuracy of model prediction, Obtain an accurate structured relationship roadmap for imaging reports
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment
[0030] Embodiment: a method for structured extraction of image reports, such as figure 1 shown, including the following steps:
[0031] S1: Obtain unstructured radiographic image text and perform preprocessing; obtain unstructured radiological image report description through the hospital RIS system or other data sources, and perform data preprocessing on the obtained long text:
[0032] 1. Remove spaces in the text;
[0033] 2. Remove the brackets and the content in the brackets. Most of the content in the brackets is content that does not require structured extraction;
[0034] 3. Unify Chinese and English punctuation marks and full-width and half-width alphanumeric letters, and delete repeated punctuation marks;
[0035] 4. Remove escape characters (such as newline characters, indentation characters, etc.);
[0036] 5. Vectorize the text and map it into the vector space.
[0037] S2: Segment the preprocessed text and then normalize it; perform word segmentation on the p...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses an image report structured extraction method. The method comprises the following steps: acquiring an unstructured radiographic image text and preprocessing the unstructured radiographic image text; carrying out word segmentation on the preprocessed text and then carrying out normalization processing; performing bert model optimization by adopting attention focus loss as an optimization function, and performing entity recognition on the normalized text on the basis of the optimized bert model; the method comprises the following steps of: extracting an entity structured relationship on the basis of an entity-extension bert model to form a structured report; according to the method, the loss function of attention-focus loss is extracted, the penalty is increased for the situation that individual word label prediction errors occur in the same entity, the loss of the labels subjected to misclassification can be increased, the loss of errors prone to classification can be reduced, and therefore the purposes of accelerating convergence of the model and improving the accuracy rate are achieved.
Description
technical field [0001] The invention relates to the technical field of text recognition and extraction, in particular to a structured extraction method for an image report. Background technique [0002] Now that medical imaging has entered the data-driven era, imaging data accounts for more than 90% of medical informatization data, and China's data volume accounts for almost 20% of the world's total. An ordinary tertiary hospital has about 3,000 image inspections a day, and 1.1 million image reports will be generated a year. This number of inspections will increase when the hospital becomes larger. Can it effectively mine the largest data in the world? Treasures of resources are of great value to the development of imaging disciplines, intelligent medical care and even the health industry. In 2007-2008, ACR and RSNA issued a series of report quality-related guidelines to encourage the construction of structured reports, improve the quality of imaging reports, and provide st...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More IPC IPC(8): G06F16/35G06F16/36G06F40/295G16H15/00
Inventor 金倍建叶金德陈集房麻元兴
Owner 浙江卡易智慧医疗科技有限公司