US 12,354,723 B2
System and method for radiology reporting
John Paulett, Berkeley, CA (US); William Boonn, Berkeley, CA (US); Woojin Kim, Berkeley, CA (US); Jeffrey Chang, Berkeley, CA (US); and Doktor Gurson, Berkeley, CA (US)
Assigned to RAD AI, Inc., San Francisco, CA (US)
Filed by RAD AI, Inc., Berkeley, CA (US)
Filed on Apr. 17, 2024, as Appl. No. 18/638,368.
Claims priority of provisional application 63/510,250, filed on Jun. 26, 2023.
Claims priority of provisional application 63/496,521, filed on Apr. 17, 2023.
Prior Publication US 2024/0347156 A1, Oct. 17, 2024
Int. Cl. G16H 15/00 (2018.01); G06F 40/289 (2020.01); G16H 30/20 (2018.01); G16H 30/40 (2018.01)
CPC G16H 15/00 (2018.01) [G16H 30/20 (2018.01); G16H 30/40 (2018.01)] 18 Claims
OG exemplary drawing
 
1. A method for automatically generating a radiology report for a radiologist, comprising:
with a computing system comprising model architecture configured for natural language processing, generating more than 75% of the radiology report upon:
receiving a first set of inputs associated with a patient at the computing system;
with an input determination model of the computing system, determining a report template based on the first set of inputs, wherein determining the report template comprises returning a ranked list of templates based upon a procedure type, an imaging modality, a patient demographic, and a study location, represented in the first set of inputs, wherein the input determination model comprises a large language model (LLM) combined with a semantic search engine;
receiving a set of unstructured findings, comprising at least one of an audio data stream and text data in free-text form, associated with the patient;
based on the set of unstructured findings and the first set of inputs, using a set of trained models of the computing system, automatically generating a set of text to populate a plurality of fields of the report template, wherein the set of text is generated with a writing style represented in word embeddings generated from historical reports of the radiologist; and
processing the report template after populating the plurality of fields, wherein processing the report template comprises correcting errors in the report template based upon inconsistencies between the report template and historical radiology report data for the radiologist in the writing style of the radiologist, upon checking word embeddings in the populated report template and in historical reports of the radiologist; and performing a billing error correction procedure, comprising:
determining predicted billing information based on a number of radiology image views provided with the set of inputs,
comparing the number of radiology image views with a field indicating number of views in the report template, and
if the number of radiology image views does not match the field, correcting at least one of the predicted billing information and the report template.