Automatically assigning medical codes using natural language processing
First Claim
1. A method of automatically interpreting a transcribed note using vector processing to generate associated codes, the method comprising the steps of:
- (a) segmenting the transcribed note into a plurality of segments;
(b) applying morphing, parsing and, semantic analysis to the segments to generate a normalized file having a standardized form with parse items, said semantic analysis comprising (1) identifying first type matches between parse items of the normalized file and a plurality of knowledge vectors, each of said knowledge vectors being manually-generated based on prior semantic knowledge and comprising data structure including one or more terms, each term having a weight based on the semantic category of the term, the weight being selected from a high weight, a low weight, and a middle weight, a high weight indicating that a term must be present in a parse item to get a match, a low weight indicating that a term, if present in a parse item, will improve a match, and a middle weight, wherein one or more of said knowledge vectors include a single term representing a numerical interval, wherein the numerical interval is represented in a vector through an interval index that map terms to semantically predefined intervals of acceptable values, the first type matches being indicative of associations to a single code, (2) generating associated codes at least on the basis of the first type matches and on the basis of natural language processing rules applied to the parse items; and
(c) outputting the generated associated codes.
7 Assignments
0 Petitions
Accused Products
Abstract
A programmed implementation for automatically assigning medical codes (including diagnosis, procedure, and level of service—i.e., evaluation and management (EM)—codes) to computer readable physician notes using natural language processing. A system allows hospitals, physician management groups, and medical billing companies to automatically determine and assign the medical codes that are the basis of reimbursement for medical services. The proposed implementation can be in place of human medical coders who otherwise would be employed to determine and assign these codes. Implementation requires only minor modifications to the traditional data-flow and allows for major improvements in throughput and timeliness of billing. Further, the present scheme enables the collection of other demographic and clinical information that resides in physician notes, but that are currently too expensive and time consuming to extract by means of a human workforce.
575 Citations
49 Claims
-
1. A method of automatically interpreting a transcribed note using vector processing to generate associated codes, the method comprising the steps of:
-
(a) segmenting the transcribed note into a plurality of segments;
(b) applying morphing, parsing and, semantic analysis to the segments to generate a normalized file having a standardized form with parse items, said semantic analysis comprising (1) identifying first type matches between parse items of the normalized file and a plurality of knowledge vectors, each of said knowledge vectors being manually-generated based on prior semantic knowledge and comprising data structure including one or more terms, each term having a weight based on the semantic category of the term, the weight being selected from a high weight, a low weight, and a middle weight, a high weight indicating that a term must be present in a parse item to get a match, a low weight indicating that a term, if present in a parse item, will improve a match, and a middle weight, wherein one or more of said knowledge vectors include a single term representing a numerical interval, wherein the numerical interval is represented in a vector through an interval index that map terms to semantically predefined intervals of acceptable values, the first type matches being indicative of associations to a single code, (2) generating associated codes at least on the basis of the first type matches and on the basis of natural language processing rules applied to the parse items; and
(c) outputting the generated associated codes. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 42, 43, 44, 45, 46, 47, 48, 49)
-
-
13. A coding system for automatically interpreting a transcribed note to generate associated codes, comprising:
-
(a) a master import/export module for retrieving the transcribed note; and
(b) a filter module for processing text associated with the transcribed note to create a normalized file; and
(c) a natural language processing (NLP) engine using vector processing to associate from the normalized file a predetermined set of codes, wherein the NLP engine comprises;
(1) a module for morphing, parsing and semantically analyzing the structures and relations within the normalized file and identifying these in the form of parse items;
(2) a first vector matching module for identifying first type matches between the parse items and a plurality of knowledge vectors, each of said knowledge vectors being manually-generated based on prior semantic knowledge and comprising a data structure including one or more terms, each term having a weight based on the semantic category of the term, the weight being selected from a high weight, a low weight, and a middle weight, a high weight indicating that a term must be present in a parse item to get a match, a low weight indicating that a term, if present in a parse item, will improve a match, and a middle weight, wherein one or more of said knowledge vectors include a single term representing a numerical interval, wherein the numerical interval is represented in a vector through an interval index that map terms to semantically predefined intervals of acceptable values, the first type matches being indicative of associations to a single code; and
(3) a code generator for generating associated codes from the predetermined set of codes on the basis of the first type matches and on the basis of rules applied to the parse items. - View Dependent Claims (14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32)
-
-
33. A computer program, stored on a computer-readable medium, for interpreting a transcribed note using vector processing to generate associated codes, the computer program comprising instructions for causing a computer to:
-
segment the transcribed note to generate a plurality of segments;
morph, parse and semantically analyze the segments to generate a normalized file having a standardized form with parse items;
identify first type matches between parse items of the normalized file and a plurality of standard knowledge vectors, each of said knowledge vectors being manually-generated based on prior semantic knowledge and comprising a data structure including one or more terms, each term having a weight based on the semantic category of the term, the weight being selected from a high weight, a low weight, and a middle weight, a high weight indicating that a term must be present in a parse item to get a match, a low weight indicating that a term, if present in a parse item, will improve a match, and a middle weight, wherein one or more of said knowledge vectors include a single term representing a numerical interval, wherein the numerical interval is represented in a vector through an interval index that map terms to semantically predefined intervals of acceptable values, the first type matches being indicative of associations to a single code;
generate associated codes at least on the basis of the first type matches and on the basis of natural language processing rules applied to the parse items; and
output the generated associated codes. - View Dependent Claims (34, 35, 36, 37, 38, 39, 40, 41)
-
Specification