# US 2021/0174923 A1 · Method, device and medium for structuring capsule endoscopy report text

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0174923 A1
- Document ID: US20210174923A1
- Status: Published application
- Kind: A1
- Application number: 17/112,971
- Filed: 2020-12-04
- Published: 2021-06-10
- Earliest claimed: 2019-12-06
- Assignees: Ankon Technologies Co., Ltd; Anx IP Holding Pte. Ltd.
- Applicants: Ankon Technologies Co., Ltd; Anx IP Holding Pte. Ltd.
- Inventors: ZHIWEI HUANG; Wenjin YUAN; Hao ZHANG; HANG ZHANG
- Web page: https://patentgrep.com/patent/US20210174923A1

## CPC

- G16H 15/00: ICT specially adapted for medical reports, e.g. generation or transmission thereof
- G06F 16/2246: {Trees, e.g. B+trees}
- G06F 40/205: Parsing
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G06F 40/30: Semantic analysis
- G06N 3/042: Knowledge-based neural networks; Logical representations of neural networks
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/09: Supervised learning
- G06N 3/096: Transfer learning
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06N 5/046: Forward inferencing; Production systems
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G16H 30/20: for handling medical images, e.g. DICOM, HL7 or PACS
- G16H 40/63: for local operation
- Y02P 90/30: Computing systems specially adapted for manufacturing
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- Y02P: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS

## Abstract

The present invention discloses a method, device and medium for structuring a capsule endoscopy report text. The method includes: annotating the report text using an annotation model; storing each named entity classification label in the report text in a hierarchical tree structure according to the annotation information to form a tree structure diagram; parsing the tree structure diagram, extracting abnormal structure data and time parameters, and storing the abnormal structure data and time parameters in a panel data structured manner to form an abnormal structure panel table and a time parameter panel table. The present invention can automatically annotate the capsule endoscopy report through the annotation model, and output parameters of different amount of information in different structures, and quantitative and accurate quality control of the capsule endoscopy process and examination results, which provide sufficient convenience for the electronic medical information of the capsule endoscopy.

## Claims

### Claim 1

A method for structuring a capsule endoscopy report text, comprising:
annotating the report text using an annotation model;
storing each named entity classification label in the report text in a hierarchical tree structure according to the annotation information to form a tree structure diagram, wherein the annotation information refers to the named category of each named entity classification label in the corresponding report text;
parsing the tree structure diagram, extracting abnormal structure data and time parameters, and storing the abnormal structure data and time parameters in a panel data structured manner to form an abnormal structure panel table and a time parameters panel table.

### Claim 2 (depends on 1)

The method of claim 1, wherein the method comprises: selectively outputting at least one of the tree structure diagrams, the abnormal structure panel table, and the time parameter panel table.

### Claim 3 (depends on 1)

The method of claim 1, wherein the method for obtaining the annotation model specifically comprises:
step M1, constructing a small neural network model using the BiLSTM+CRF structure and constructing a large neural network model using the BERT structure, wherein the initial small model and the large model have the same named entity division rules, and pre-annotating a current set of report texts using the small model and the large model, respectively;
step M2, reviewing and correcting the current report text annotated by the large model to form a revised report text;
step M3, verifying the small model with the revised report text, and obtaining the log-likelihood loss function corresponding to the small model;
step M4, revising the named entity division rules corresponding to the small model with the currently obtained revised report text and training the small model, and using the trained small model as an annotation model, when the log-likelihood loss function is not greater than the preset first function value; proceeding to step M5, when the log-likelihood loss function is greater than the preset first function value;
step M5, revising the named entity division rules corresponding to the small model and the large model with the currently obtained revised report text, and training the large model and the small model, and pre-annotating the next set of report texts using the trained small model and large model at the same time, wherein the quantity of any next set of report texts is greater than the quantity of previous set of report texts; and executing step M2 in a loop until the trained small model becomes an annotation model.

### Claim 4 (depends on 1)

The method of claim 1, wherein the step “annotating the report text using an annotation model” specifically comprises:
pre-processing the report text to organize the report text into a recognizable report text;
parsing the recognizable report text, and recognizing each named entity classification label in the report text according to the named entity division rules;
traversing the recognizable report text, and annotating each named entity classification label and the text that is different from the named entity classification label using the BIO labeling method.

### Claim 5 (depends on 4)

The method of claim 4, wherein the step “pre-processing the report text” specifically comprises:
searching the original report text based on pre-set rules, correcting misspellings, mixed Chinese and English characters, mixed upper- and lower-case characters, punctuation errors, abnormal numerical descriptions, irrelevant characters, and irregular descriptions in the original report text.

### Claim 6 (depends on 4)

The method of claim 4, wherein before the step “parsing the recognizable report text, and recognizing each named entity classification label in the report text according to the named entity division rules”, the method further comprises: establishing a named entity division rule,
wherein the named entity division rule comprises:
establishing a corresponding relationship between the named entity classification label and the annotation information, the annotation information comprising: organ, description information corresponding to the organ, abnormal structure corresponding to the organ, description parameter corresponding to the abnormal structure, and time parameter identification.

### Claim 7 (depends on 4)

The method of claim 4, wherein the step “traversing the recognizable report text, and annotating each named entity classification label and the text that is different from the named entity classification label using the BIO labeling method” specifically comprises: annotating the initial character of each named entity classification label using a combination of a first prefix and annotation information, annotating the non-initial characters of each named entity classification label using a combination of a second prefix and annotation information, and annotating the text without annotation information with a third prefix, wherein the first prefix, the second prefix, and the third prefix are all different.

### Claim 8 (depends on 3)

The method of claim 3, wherein in the process of pre-annotating each set of report texts, the method further comprises:
segmenting each report text into multiple short sentences through punctuation, processing the same short sentences by deduplication and retaining one for pre-annotation, review and correction.

### Claim 9

An electronic device, comprising a memory and a processor, wherein the memory stores computer programs that can run on the processor
and the processor executes the computer programs to implement the steps of a method for structuring a capsule endoscopy report text,
wherein the method comprises:
annotating the report text using an annotation model;
storing each named entity classification label in the report text in a hierarchical tree structure according to the annotation information to form a tree structure diagram, wherein the annotation information refers to the named category of each named entity classification label in the corresponding report text;
parsing the tree structure diagram, extracting abnormal structure data and time parameters, and storing the abnormal structure data and time parameters in a panel data structured manner to form an abnormal structure panel table and a time parameters panel table.

### Claim 10

A computer-readable storage medium storing computer programs,
wherein the computer programs can be executed by the processor to implement the steps of a method for structuring a capsule endoscopy report text,
wherein the method comprises:
annotating the report text using an annotation model;
storing each named entity classification label in the report text in a hierarchical tree structure according to the annotation information to form a tree structure diagram, wherein the annotation information refers to the named category of each named entity classification label in the corresponding report text;
parsing the tree structure diagram, extracting abnormal structure data and time parameters, and storing the abnormal structure data and time parameters in a panel data structured manner to form an abnormal structure panel table and a time parameters panel table.
