TW202226260A - Smart image diagnosis report translation method - Google Patents

Abstract

A smart image diagnosis report translation method and system thereof, wherein the method includes: performing a training image input step; performing a text symbol screening step to capture text symbol information on the those training image diagnosis report; performing a format standardization step, and generating a standardized diagnosis report format; executing an convert-expected image input step and executing a standardized conversion step, and then compare the text symbol information to be converted with the standardized diagnostic report format to obtain a standardized diagnostic conversion report. A smart image diagnosis report translation system, including: an image capture module, an image analysis module, a text symbol processing module, a standardized format generation module, and a diagnosis report conversion module, and then compare and analyze the convert-expected diagnosis report image with the standardized diagnostic report format to generate a standardized diagnostic conversion report.

Description

Intelligent imaging diagnosis report translation method and system

The present invention relates to a diagnostic report translation method and system, in particular to a method of using paper diagnostic report images to capture text information through image recognition, and using neural network learning to convert paper diagnostic reports into standard digital format. Intelligent imaging diagnosis report translation method and system.

At present, the treatment tracking method of alcoholics is still mainly in the way of visiting medical institutions to be consulted by medical staff, which is very inconvenient and inefficient for patients who need long-term tracking. It is also a time-consuming task to electronically record the alcohol detection data at the next visit. Therefore, it is necessary to develop a real-time and remote monitoring alcohol detection record system to assist patients, family members and medical units. necessary.

Known Technology The Republic of China Patent Application No. 100130595 relates to a computerized automatic human nailfold microcirculation analysis system, which is used to capture pathological data to provide a basis for doctors to diagnose, which includes: a microcirculation image sequence input device; A living body shaking stabilization device; a nailfold microvessel identification device for image processing in spatial and temporal domains; a microcirculation static and dynamic pathological feature analysis and pathological data generating device; and a three-dimensional stereoscopic dynamic blood flow rendering platform.

Conventional Technology The Republic of China Patent No. 594532 provides a medical data system for processing various data formats and data sources and uses the system to effectively complete the medical records of a patient scattered in various hospitals or departments A method of integrating transmission on a user end.

Known Technology Republic of China Patent No. I687938 provides a method for establishing an electronic medical record, including the following steps: converting a paper medical record into an electronic image, storing the electronic image in a database, The electronic image is opened in the device to form an electronic medical record image with a layer, so that a user can further input a character, handwrite a character, draw a pattern, or paste a digital media file on the electronic medical record image, to create a medical information, and store the electronic medical record image in the database.

However, none of the above-mentioned prior art discloses the function of interpreting and analyzing the text content on the written diagnosis report, and re-translating and converting the content of the written diagnosis report into a digital standardized format by means of neural network learning. It is convenient for the record as a digitized standard format and the readers of the diagnosis report to understand the content of the diagnosis report more easily. Therefore, it is necessary to improve the conventional technology.

An object of the present invention is to provide an intelligent image diagnosis report translation method and system, which translates the content of the paper diagnosis report by means of image recognition, and filters the content.

Another object of the present invention is to provide a method and system for translating an intelligent image diagnosis report, which translates the content of a paper diagnosis report into a standard structured format by means of neural network learning.

Another object of the present invention is to provide an intelligent image diagnosis report translation method and system, which translates the content of a paper diagnosis report into a standard structured digital diagnosis report.

Another object of the present invention is to provide an intelligent image diagnosis report translation method and system, which can directly translate the diagnosis result into a standard structured digital diagnosis report.

In order to achieve the above and other objects, the intelligent image diagnosis report translation method of the present invention includes: performing a training image input step to provide a plurality of training image diagnosis reports; Text symbol information, and perform sentence segmentation on the text symbol information to obtain plural initial words; perform a format standardization step, carry out feature classification and formatting of the plural initial words, and generate a standardized diagnosis report format; perform an image input to be converted Steps: inputting an image to be converted into a diagnostic report, performing an information capture step on the image to be converted into a diagnostic report to obtain a text symbol information to be converted; and performing a standardization conversion step of combining the text symbol information to be converted with the text symbol information to be converted Standardized diagnostic report formats are compared to obtain a standardized diagnostic conversion report.

In order to achieve the above and other objects, the intelligent image diagnosis report translation system of the present invention includes: an image capture module for capturing a plurality of training image diagnosis reports and generating a plurality of training target images; an image analysis module for using to analyze the complex training target image and generate a text symbol database; a text symbol processing module analyzes the text symbol database and generates a complex initial vocabulary; a standardization format generating module converts the complex initial vocabulary into a standardization process A diagnosis report format; and a diagnosis report conversion module, which generates a standardized diagnosis conversion report by comparing and analyzing an image to be converted into a diagnosis report and the standardized diagnosis report format.

In some embodiments of the present invention, wherein the word symbol screening step package further includes a word filtering and sorting step, the word filtering and sorting step is to filter out unnecessary words, symbols and numbers from the plural initial words, and Perform word frequency calculation and word sorting on the filtered words.

In some embodiments of the present invention, the step of filtering and sorting words further includes a step of filtering key words, and by providing a key word data table, the words filtered by the filtering and sorting step and the key words After comparing the data tables, the plural key words are screened out.

In some embodiments of the present invention, after the character symbol screening step, a weight adjustment step is further included. By providing a weight adjustment data table, the plural key words are adjusted to each key word according to the weight adjustment data table. Classify and give weight adjustment respectively.

In some embodiments of the present invention, the text symbol processing module further includes a vocabulary filtering and sorting unit for filtering and sorting the text symbol data.

In some embodiments of the present invention, the character symbol processing module further includes a key word screening unit, and the key word screening unit is used for comparing through a key word data table, and after screening and sorting Plural key words are filtered out from the text symbol data.

In some embodiments of the present invention, the character symbol processing module further includes a weight adjustment unit, and the weight adjustment unit adjusts the weight according to the importance of words.

In some embodiments of the present invention, the diagnostic report conversion module further includes a quick conversion interface unit, the quick conversion interface unit provides the user with inputting the diagnostic result, and the quick conversion interface unit converts the diagnostic result to output into a standardized diagnostic conversion report.

FIG. 1 is a flowchart of an embodiment of a method for translating an intelligent image diagnosis report according to the present invention. Please refer to FIG. 1 . The intelligent image diagnosis report translation method of the present invention includes: executing a training image input step S0 to provide a plurality of training image diagnosis reports, and the multiple training image diagnosis reports are captured from traditional paper diagnosis reports by scanning, reproducing or other images. Obtain image files of paper diagnostic reports, such as Magnetic Resonance Imaging (MRI) or Computed Tomography (CT) and other image diagnostic reports.

A text symbol screening step S1 is performed to capture text symbol information on the plural training image diagnosis report, and perform sentence segmentation on the text symbol information to obtain plural initial words. Since these paper image diagnosis reports record the diagnosis content of each doctor for different patients, and each doctor writes in a different format, there is no rule in the content of each image diagnosis report. Therefore, through the training image input step S0 converts the content of the image diagnosis report into image file data, which is helpful for analyzing the report content. In the text symbol screening step S1, the image content of the complex training image diagnosis report is subjected to image analysis and identification such as sentences, words and symbols, etc. The image data can be converted into digital text and symbol data, which is convenient for computers and software programs to read, that is, the text on the paper report is converted into image format and then converted into digital text format data.

A format standardization step S2 is performed to classify and format the plural initial words to generate a standardized diagnosis report format. In the format standardization step S2, the plural initial words are classified by noun features, such as: pathology, affected part, affected part, development and change, etc., and a formatting model is given, and the formatting model can be formulated according to requirements. Without limitation, all the plural initial words and the formatted model obtained from the plural training image diagnostic report provided in step S0 are input from the training image to perform machine learning, and the standardized diagnostic report format is generated by a neural network training method. , the standardized diagnostic report format is used as a reference sample for subsequent image diagnostic report file conversion.

Execute an inputting step S3 of an image to be converted, input an image to be converted into a diagnostic report, perform an information extraction step S30 on the image to be converted into a diagnostic report, and obtain a text symbol information to be converted, and the information retrieval step S30 uses Perform image analysis of sentences, words, and symbols on the image content of the diagnostic report to be converted, identify the analyzed content, and convert the image format to obtain a text format for subsequent comparison and conversion.

and executing a standardized conversion step S4, comparing the text symbol information to be converted with the standardized diagnosis report format to obtain a standardized diagnosis conversion report. In the standardization conversion step S4, the word symbol information to be converted is filled in the standardized diagnosis conversion report in sequence according to the standardized diagnosis report format, so that it becomes an easy-to-read and computer-readable diagnosis report, such as radiation If physicians or other physicians want to view the diagnosis report, the hospital's medical record system will read the standard structured diagnosis report file for the physician to read, which can avoid the follow-up related personnel being caused by different physicians due to different usage habits of filling out the diagnosis report. Inconvenience when reading diagnostic reports. FIG. 2 is a flowchart of another embodiment of the method for translating an intelligent image diagnosis report according to the present invention. Please refer to FIG. 2 . Preferably, the word symbol screening step S1 package further includes a word filtering and sorting step S10, the word filtering and sorting step S10 is to filter out unnecessary words, symbols and numbers from the plural initial words, and filter the filtered words. The vocabulary frequency is calculated and the vocabulary is sorted. The vocabulary frequency calculation is to calculate the number of occurrences of the same vocabulary, and arrange each vocabulary in order of frequency of occurrence, which is more conducive to the standardization of the image diagnosis report format reference. This step is mainly: To extract single-word (OneGram), two-word (BiGram) and three-word (TriGram), and sort by word frequency, and finally select the available words.

FIG. 3 is a flowchart of another embodiment of the method for translating an intelligent image diagnosis report according to the present invention. Please refer to FIG. 3 . Preferably, the word filtering and sorting step S10 further includes a key word screening step S101, by providing a key word data table, the words filtered by the word filtering and sorting step S10 are compared with the key word data table. After matching, a plurality of key words are screened out. The key word data table is obtained by comparing and filtering words whose occurrences exceed a threshold value and are not redundant words, and then compare and filter them with a word filter table, wherein the threshold Preferably, the value can be adjusted according to the verification accuracy rate through rolling machine learning. For example, referring to the result of the format standardization step S2, the threshold value is preferably initially set to 3~7. In this embodiment, the threshold value is 5, and the The setting of the threshold value can improve the accuracy of the plural key words. The so-called redundant words are words such as it, a, is, etc. The word selection table is a vocabulary table provided by professionals in advance.

，0 ＜ i ≤ k，則: 步驟1. 計算詞彙集詞彙總次數F。 步驟2. 從C找出含有t _i的c與對應的f。 步驟3. 對每一個r，找出所有含r的c，並加總對應的f/F設為s。 步驟4. 則依據s，排序r，挑出r，當作預測結果。 FIG. 4 is a flowchart of another embodiment of the method for translating an intelligent image diagnosis report according to the present invention. Please refer to FIG. 4 . Preferably, after the character symbol screening step S1, a weight adjustment step S11 is further included, by providing a weight adjustment data table, the plurality of key words are classified according to the weight adjustment data table and each key word is classified and given respectively. Weight adjustment, after the machine learning, among the associated words, the better weight is obtained by the correlation performance of the words. Suppose there is a word set T={t1, t2, …., tn} and a vocabulary set R={r1, r2, . c _(m*n) }, where c represents any combination of t and r c={t, r}. If the number of occurrences of tn and rm together is expressed as f, when k words are taken out of a sentence,

, 0 < i ≤ k, then: Step 1. Calculate the total number of words F in the vocabulary set. Step 2. Find the c containing t _i and the corresponding f from C. Step 3. For each r, find all c containing r, and sum up the corresponding f/F as s. Step 4. According to s, sort r, pick out r, and use it as the prediction result.

For example: if A, Y, Z are words selected in the character symbol screening step S1, B, C, X are words listed in the weight adjustment data table, and (A, B) appear together 5 times, (A, X) appear together 3 times, (Y, B) appear together 2 times, (Z, C) appear together 8 times, assuming the total vocabulary set is 100, where the total vocabulary set refers to the symbol in the text The number of words that appear together in the training data between the words screened out in the screening step S1 and the words listed in the weight adjustment data table, the weight of the (A, B) word set is 0.05, accounting for 100 words in the total word set. The ratio is 5/100. Similarly, the weight of the (A, X) vocabulary set is 0.03, the weight of the (Y, B) vocabulary set is 0.02, and the weight of the (Z, C) vocabulary set is 0.08. If A is found in the test data, B will be given priority as the final prediction result. From another perspective, if there is a test data to find A, Y and Z, then B can get the weight of (0.05+0.02=0.07) , C can get a weight of 0.08, X can get a weight of 0.03, so C will be given priority as the final prediction result, that is, as a vocabulary in the standard format, which is more convenient and practical. The standardized diagnosis report format, with the gradual increase of data, the system will use this format to carry out machine learning, and continuously and dynamically adjust the weight according to the proportion of the total vocabulary occurrences to generate the aforementioned standardized diagnosis report format, so that the standardized diagnosis report The format tends to be more correct.

FIG. 5 is a schematic diagram of a system structure of the intelligent image diagnosis report translation system of the present invention, please refer to FIG. 5 . The intelligent image diagnosis report translation system of the present invention includes: an image capture module 10 , an image analysis module 20 , a text symbol processing module 30 , a standardized format generation module 40 and a diagnosis report conversion module 50 , the image capturing module 10 is used for capturing multiple training image diagnosis reports and generating multiple training target images. The multiple training image diagnosis reports can obtain multiple training target images through image scanning, and the multiple training target images are the diagnosis Video file of the report.

The image analysis module 20 is used to analyze the plural training target images and generate a text symbol database, the text symbol database stores the text data in the plural training target images, and the image analysis module 20 converts the text images into Plain text format for subsequent analysis.

The word symbol processing module 30 analyzes the word symbol database and generates plural initial words, wherein the plural initial words are sentences, words and symbols in the word symbol database by the word symbol processing module 30 and so on for interpretation.

The standardized format generating module 40 converts the plural initial words into a standardized diagnosis report format. The standardized format generating module 40 performs machine learning on all the obtained plural initial words and a formatted model, through, for example, a neural network The training method generates a standardized diagnostic report format, and the standardized diagnostic report format can be adjusted and revised according to actual needs, so that it can be used as a reference for the subsequent conversion of paper-based traditional imaging diagnostic reports into electronic diagnostic reports.

The diagnostic report conversion module 50 compares and analyzes an image to be converted into a diagnostic report with the standardized diagnostic report format to generate a standardized diagnostic report format, because the standardized diagnostic report format is based on the previous image diagnostic report through machine learning In the digital standard diagnostic report format generated by training, the diagnostic report conversion module 50 can convert the input image to be converted into a diagnostic report after analyzing its content and convert it into a standardized diagnostic conversion report according to the standardized diagnostic report format.

FIG. 6 is a system architecture diagram of another embodiment of the intelligent image diagnosis report translation system of the present invention, please refer to FIG. 6 . Preferably, the text symbol processing module 30 further includes a vocabulary filtering and sorting unit 31 for filtering and sorting the text symbol data. The word symbol processing module filters out unnecessary words, symbols and numbers for the plural initial words, and performs word frequency calculation and word order on the filtered words.

Please continue to refer to Figure 6. Preferably, the text symbol processing module 30 further includes a key word filtering unit 32, and the key word filtering unit 32 is used for comparing the text symbol data after filtering and sorting through a key word data table. Filter out the plural key words. Among them, the key vocabulary information table includes proper nouns such as: (Symptoms: ground glass opacity), (Organ: Lung), (Position: LUL), (Progressive: regressive change) and (Enhancement: Early Enhanced).

Please continue to refer to Figure 6. Preferably, the character symbol processing module 30 further includes a weight adjustment unit 33 , the weight adjustment unit adjusts the weight according to the importance of the vocabulary to generate a more convenient and practical standardized diagnostic report format.

Please continue to refer to Figure 6. Preferably, the diagnostic report conversion module further includes a quick conversion interface unit 51, the quick conversion interface unit 51 provides the user with inputting the diagnostic result, and the quick conversion interface unit 51 converts the diagnostic result into a standardized diagnostic conversion output. Report. Among them, the quick conversion interface unit 51 is a cloud interface system in this embodiment, for example, it can provide a radiologist to log in to the cloud interface system when writing an image diagnosis report, by selecting a click mode or a menu, etc. way to quickly and easily complete a CT image diagnosis report, and this CT image diagnosis report is an image diagnosis report created in a standardized diagnosis report format, and if the radiologist or attending physician wants to view the diagnosis report, the cloud interface The system will read the standardized diagnostic report file and display the diagnostic report in a web interface.

To sum up, the present invention uses traditional medical imaging diagnosis reports to check and parse one by one. Since medical imaging diagnosis reports are all written by doctors in natural language, the content is mostly simple sentences or compound sentences composed of English words and medical professional vocabulary. , which is then structured into a paragraph by multiple sentences, describing or explaining its findings through video inspection in detail. The invention can convert the historical diagnosis report data stored in the hospital for a long time and cannot be reused into a standard structured diagnosis that can be accurately identified, read, processed and utilized by a computer information system, and can be exchanged between different information systems. Report. Therefore, the intelligent image diagnosis report translation system of the present invention reads many traditional unstructured CT diagnosis reports as a training data set for artificial intelligence machine learning, and uses the information retrieval (Information Retrieval) technology to retrieve the content of the report from the report. Produce information such as lexical feature values and lexical frequencies, and then use machine learning methods to train the obtained keyword lexical feature values to confirm the relevance and percentage of keyword occurrences, and then perform correlation analysis with the text, and finally the domain experts . Specialists provide professional judgment and advice, and then summarize and define the standard structured CT diagnostic report format for this type of disease. After that, the system can extract information from the untrained unstructured CT radiological diagnosis report and combine it with the standard structure. Compare the CT diagnostic report format, and automatically translate the CT diagnostic report in a standard structured format.

The above-mentioned embodiments are only to illustrate the technical ideas and features of the present invention, and the purpose is to enable those who are familiar with the art to understand the content of the present invention and implement it accordingly, and should not limit the patent scope of the present invention. , all equivalent changes or modifications made according to the spirit of the present invention and the contents of the description shall be covered within the scope of the patent of the present invention.

S0: training image input step S1: Text symbol filtering step S10: Vocabulary filtering and sorting steps S101: Screening steps for key words S11: Weight Adjustment Step S2: Format Standardization Step S3: Input steps to convert images S30: Information retrieval steps S4: Standardized transformation steps 10: Image capture module 20: Image Analysis Module 30: Text Symbol Processing Module 40: Standardized format generation module 50: Diagnostic report conversion module

FIG. 1 is a flowchart of an embodiment of a method for translating an intelligent image diagnosis report according to the present invention; 2 is a flowchart of another embodiment of the method for translating an intelligent image diagnosis report according to the present invention; 3 is a flowchart of another embodiment of the method for translating an intelligent image diagnosis report of the present invention; FIG. 4 is a flowchart of another embodiment of the method for translating an intelligent image diagnosis report according to the present invention; 5 is a schematic diagram of a system architecture of the intelligent image diagnosis report translation system of the present invention; FIG. 6 is a system architecture diagram of another embodiment of the intelligent image diagnosis report translation system of the present invention.

S0: training image input step

S1: Text symbol filtering step

S2: Format Standardization Step

S3: Input steps to convert images

S4: Standardized transformation steps

Claims (10)

An intelligent image diagnosis report translation method, comprising: Performing a training image input step to provide a diagnostic report of plural training images; performing a text symbol screening step, retrieving text symbol information on the plural training image diagnosis report, and performing sentence segmentation on the text symbol information to obtain plural initial words; Performing a format standardization step, classifying and formatting the plural initial words, and generating a standardized diagnosis report format; performing an inputting step of an image to be converted, inputting an image to be converted into a diagnostic report, and performing an information capture step on the image to be converted into a diagnostic report to obtain a text symbol information to be converted; and A standardized conversion step is performed to compare the text symbol information to be converted with the standardized diagnosis report format to obtain a standardized diagnosis conversion report. The method for translating an intelligent image diagnosis report with image recognition according to claim 1, wherein the character symbol filtering step further comprises a vocabulary filtering and sorting step, and the vocabulary filtering and sorting step is to filter out the plurality of initial words Unnecessary words, symbols and numbers are removed, and the filtered vocabulary is subjected to vocabulary frequency calculation and vocabulary sorting. The method for translating an intelligent image diagnosis report with image recognition according to claim 2, wherein the step of filtering and sorting words further comprises a step of filtering key words, and filtering the words by providing a key word data table After the words filtered by the sorting step are compared with the key word data table, plural key words are filtered out. The method for translating an intelligent image diagnosis report with image recognition according to claim 1, wherein after the character symbol screening step, a weight adjustment step is further included, by providing a weight adjustment data table, the plurality of key words According to the weight adjustment data table, each key word is classified and given weight adjustment respectively, wherein the weight adjustment step is provided with a threshold value, and the threshold value can be adjusted through rolling machine learning. The method for translating an intelligent image diagnosis report with image recognition according to claim 4, wherein the initial value of the threshold value is 3-7. An intelligent image diagnosis report translation system using the intelligent image diagnosis report translation method as claimed in item 1, comprising: an image capturing module for capturing multiple training image diagnostic reports and generating multiple training target images; an image analysis module for analyzing the plurality of training target images and generating a text symbol database; A word symbol processing module, which analyzes the word symbol database and generates plural initial words; a standardized format generation module that converts the plural initial words into a standardized diagnostic report format; and A diagnosis report conversion module generates a standardized diagnosis conversion report by comparing and analyzing an image to be converted into a diagnosis report with the standardized diagnosis report format. The intelligent image diagnosis report translation system according to claim 6, wherein the text symbol processing module further comprises a vocabulary filtering and sorting unit for filtering and sorting the text symbol data. The intelligent image diagnosis report translation system according to claim 6, wherein the character symbol processing module further comprises a key word screening unit, and the key word screening unit is used for comparing through a key word data table, Plural key words are filtered out from the filtered and sorted text symbol data. The intelligent image diagnosis report translation system according to claim 6, wherein the character symbol processing module further comprises a weight adjustment unit, and the weight adjustment unit adjusts the weight according to the importance of words. According to the intelligent image diagnosis report translation system according to claim 6, the diagnosis report conversion module further comprises a quick conversion interface unit, the quick conversion interface unit provides the user with inputting the diagnosis result, and the quick conversion interface unit converts The diagnostic result is converted and output into a standardized diagnostic conversion report.

Images