Automated identification of salient finding codes in structured and narrative reports
In a tool for assisting in summarizing salient medical report findings, finding tokens ( 66 ) representing findings are extracted from a current medical report, and a salience value is computed for each such finding token based on statistics for the finding token in a reference database ( 32 ) including a count of reference medical reports of the containing the finding token in a report summary section and a count of reference medical reports containing the finding token anywhere in the report. A ranking or sub-set of findings extracted from the current medical report is displayed. The displayed findings may be ranked, or the sub-set chosen, by salience value. To generate the reference database, for each reference medical report ( 42 ) finding tokens ( 46 ) are extracted and it is determined whether each such finding token is salient based on whether the finding token is extracted from a report summary section of the reference report.
1. A medical reporting device comprising:
a training computer programmed to perform a training process including, for each reference medical report of a reference database (i) matching text of the reference medical report to finding templates of a set of finding templates to extract finding tokens representing findings from the reference medical report and (ii) determining whether each finding token extracted from the reference medical report is salient in the reference medical report; and
a report entry computer including a display and one or more user interface devices, the report entry computer programmed to:
receive user entry of a current medical report via the one or more user interface devices and at least a portion of the current medical report by filling in a structured medical report form is user selection via the one or more user interface devices of standard finding codes of a set of standard codes;
perform a salient findings identification process including: (I) matching text of the current medical report to finding templates of the set of finding templates to extract finding tokens representing findings from the current medical report and matching text of the current medical report to standard finding codes of the set of standard finding codes and (II) computing a salience value for each finding token extracted from the current medical report based on statistics generated by the training process including a count of reference medical reports from which the finding token was extracted and determined to be salient and a count of reference medical reports from which the finding token was extracted; and
present, on the display, a sub-set or all findings extracted from the current medical report ranked by salience value.
2. The medical reporting device of claim 1 wherein the operation (ii) comprises:
determining whether each finding token extracted from a report summary section of the reference medical report is salient in the reference medical report, the finding token being not salient otherwise.
3. The medical reporting device of claim 1 wherein the operation (ii) comprises:
receiving from a user an indication of whether each finding token extracted from the reference medical report is salient in the reference medical report.
4. The medical reporting device of claim 1 wherein each finding template of the set of finding templates comprises a set of semantic fields wherein each semantic field is associated to a semantic category of a medical ontology.
5. The medical reporting device of claim 4 wherein semantic fields of the finding templates of the set of finding templates include at least: a semantic field associated to an anatomy category of the medical ontology; a semantic field associated to a diagnosis category of the medical ontology; and a semantic field associated to a severity category of the medical ontology.
6. The medical reporting device of claim 1 wherein the matching operations (i) and (I) comprise:
extracting words or phrases of a medical ontology from the reference or current medical report; and
identifying combinations of the extracted words or phrases matching finding templates of the set of finding templates to extract finding tokens representing findings from the reference or current medical report.
7. The medical reporting device of claim 1 wherein, in the operation (II), the counts are limited to reference medical reports having a predefined set of finding tokens also extracted from the current medical report.
8. The medical reporting device of claim 1 wherein, in the operation (II), the counts are limited to reference medical reports having a predefined context, the current medical report also having the predefined context.
9. The medical reporting device of 1 wherein the report entry computer is programmed to present said sub-set or all findings by presenting text of the current medical report representing each finding.
10. The medical reporting device of claim 1 wherein the training computer and the report entry computer are a single computer.
11. A non-transitory storage medium storing instructions readable and executable by one or more computers to perform a medical reporting method comprising:
receive user entry of a current medical report via one or more user interface devices and at least a portion of the current medical report in a structured medical report form including user selection via the one or more user interface devices of standard finding codes of a set of standard finding codes:
for each reference medical report of a reference database: (i) extracting finding tokens representing findings from the reference medical report and (ii) determining whether each finding token extracted from the reference medical report is salient in the reference medical report based on whether the finding token is extracted from a report summary section;
extracting finding tokens representing findings from a current medical report;
computing a salience value for each finding token extracted from the current medical report based on statistics for the finding token in the reference database including a count of reference medical reports from which the finding token was extracted and determined to be salient and a count of reference medical reports from which the finding token was extracted; and
presenting, on a display, at least one of all findings represented by finding tokens extracted from the current medical report ranked by salience value and a sub-set of all findings represented by finding tokens extracted from the current medical report having highest salience value.
12. The non-transitory storage medium of claim 11 wherein the extracting of finding tokens from the reference or current medical report includes:
extracting words or phrases of a medical ontology from the reference or current medical report; and
identifying combinations of the extracted words or phrases matching finding templates of a set of finding templates wherein the identified combinations form the finding tokens representing findings from the reference or current medical report.
13. The non-transitory storage medium of claim 11 wherein the counts are limited to reference medical reports having a predefined set of finding tokens which are also extracted from the current medical report.
14. The non-transitory storage medium of claim 11 wherein the counts are limited to reference medical reports having a predefined context, the current medical report also having the predefined context.
15. The non-transitory storage medium of claim 11 wherein the presenting includes presenting said all findings or said sub-set of all findings by presenting text of the current medical report representing each finding.