US 11,861,314 B2
Extracting clinical follow-ups from discharge summaries
Yada Pruksachatkun, New York, NY (US); Sean Adler, New York, NY (US); Thomas Gregory McKelvey, Jr., New York, NY (US); Jordan Louis Swartz, New York, NY (US); Hui Dai, New York, NY (US); Yi Yang, Long Island City, NY (US); David Sontag, Brookline, MA (US); and Jennifer Marie Seale, Austin, TX (US)
Assigned to ASAPP, INC., New York, NY (US)
Filed by ASAPP, INC., New York, NY (US)
Filed on Apr. 2, 2021, as Appl. No. 17/221,485.
Claims priority of provisional application 63/004,901, filed on Apr. 3, 2020.
Prior Publication US 2021/0312128 A1, Oct. 7, 2021
Int. Cl. G06F 40/30 (2020.01); G06F 40/284 (2020.01); G16H 10/60 (2018.01); G16H 10/40 (2018.01); G06N 3/04 (2023.01); G06F 18/2431 (2023.01); G06V 20/62 (2022.01)
CPC G06F 40/30 (2020.01) [G06F 18/2431 (2023.01); G06F 40/284 (2020.01); G06N 3/04 (2013.01); G16H 10/40 (2018.01); G16H 10/60 (2018.01); G06V 20/62 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A method for identifying actionable content from a medical record, the method comprising:
retrieving a focus sentence from the medical record;
retrieving contextual data for the focus sentence from the medical record, wherein the contextual data includes words before and after the focus sentence in the medical record;
tokenizing the focus sentence and the contextual data;
determining embeddings for the words of the tokenized focus sentence and the contextual data;
processing the embeddings with a word embedding model to determined contextualized embeddings of the words in the focus sentence;
processing the contextualized embeddings with a sentence embedding model to determine a sentence embedding vector for the focus sentence;
determining a position of the focus sentence in the medical record;
processing the sentence embedding vector and the position with a multi-label classifier to determine a multi-label score vector, wherein each element of the multi-label score vector is associated with a different type of actionable content;
assigning at least one type of actionable content to the focus sentence based on values of the elements of the score vector, and
emphasizing the focus sentence in the medical record according to the assigned type of actionable content.