Extraction of PERSON and LOCATION from Clinical Study Report, using cleanNLP with Stanford coreNLP backend

Ken Ikeda
Open access to original clinical study reports is important for health science development. Then, the challenge is to efficiently detect personal information , in order to prserve the anonymity of study participants. Such information includes, but not limited to, person names, addresses, etc. The data like email addresses and phone numbers can be easily detected based on their string patterns by means of base R functions and the R packages like stringi and stringr. So I have been seeking methods to detect ADDRESS, NAME and ORGANIZATION in texts.
This is a trial to detect ADDRESS and NAME in medical reports, using the cleanNLP package with Stanford coreNLP engine. Stanford coreNLP looked to have a better capability in named entity recognition than openNLP with “openNLPmodels.en”" but this remains to be studied. Of note is that when I ran the same scripts on an Windows 8 machine, encountered was an encoding error (use ASCII/UTF-8!) at the annotation step, cnlp_annotate(). This did not happen in R 3.4.3 on Mac OS High Sierra.
The following clinical study report was used: “http://www.maps.org/research-archive/mdma/MP-2_CSR_FINAL_15Sep11.pdf”

Install the cleanNLP package and the coreNLP model

You need to run the following commands

library(markdown)
library(dplyr)
library(pdftools)
library(cleanNLP)
library(stringr)
library(knitr)
#install.packages("cleanNLP", dependencies = T)
#library(cleanNLP)
#cnlp_download_corenlp() # This took a few minutes

Read a sample pdf report and annotate it

The csr was a vector of characters (62 pages).
You must execute the cnlp_init_corenlp() function before analyzing new text data.
anno_level = 2 has named-entity recognition, if it is set to 3 or higher, more returns.

#Read a sample Clinical Study Report
csr <- pdf_text("http://www.maps.org/research-archive/mdma/MP-2_CSR_FINAL_15Sep11.pdf")
#Everytime you must run the FUNCTION cnlp_init_corenlp(), this needs take few seconds but faster in the subsequent
library(rJava)
cnlp_init_corenlp(language = "en", anno_level = 2, mem = "6g")
#Annotate the Clinical Study Report
annotated <- cnlp_annotate(csr, as_strings = TRUE, backend = "coreNLP")

Locate Data of interests by Entity Type

Entities available are:
“LOCATION”, “PERSON”, “ORGANIZATION” ,“MONEY”, “PERCENT”,“DATE”, and “TIME”.

annotated$entity %>% 
    filter(entity_type %in% c("PERSON", "LOCATION")) %>%
    select(id, entity_type, entity) %>% 
    mutate(page=as.integer(str_extract(id, "\\d+"))) %>% 
    select(page, entity_type, entity) %>% 
    arrange(page) %>% kable(format="markdown")

page	entity_type	entity
1	LOCATION	309 Cedar Street
1	LOCATION	# 2323 Santa Cruz
1	PERSON	Rick Doblin
1	PERSON	Peter Oehen
2	PERSON	Peter Oehen
2	PERSON	Verena Widmer
2	PERSON	Rafael Traber
2	LOCATION	Switzerland
4	PERSON	Oehen
10	LOCATION	U.S.
13	LOCATION	Canton
13	LOCATION	Aargau
13	LOCATION	Solothurn
13	LOCATION	Switzerland
13	LOCATION	Helsinki
13	PERSON	Peter Oehen
13	PERSON	Oehen
13	PERSON	Verena Widmer
13	PERSON	Oehen
13	LOCATION	Switzerland
13	PERSON	Oehen
13	PERSON	Rafael Traber
13	PERSON	Peter Oehen
13	PERSON	Verena Widmer
13	PERSON	Rafael Traber
13	PERSON	Rudolf Brenneisen
13	PERSON	Michael Mithoefer
13	LOCATION	Ilsa Jerome
13	PERSON	Christoph Kopp
13	LOCATION	U.S.
13	LOCATION	Europe
13	LOCATION	Spain
14	LOCATION	U.S
14	PERSON	Stanislav Grof
14	PERSON	Holotropic Breathwork
17	PERSON	Oehen
19	LOCATION	Switzerland
19	PERSON	Franz Vollenweider
19	LOCATION	Switzerland
19	PERSON	Rudolf Brenneisen
19	LOCATION	Switzerland
19	PERSON	Brenneisen
19	PERSON	Laboratory Bichsel
19	LOCATION	Interlaken
19	LOCATION	Switzerland
20	PERSON	M. Collenberg
20	LOCATION	U.S.
20	PERSON	Charles Grob
28	PERSON	Wilcoxon
34	PERSON	Mauchly
36	PERSON	Mauchly
37	PERSON	Mauchly
37	PERSON	Mauchly
37	PERSON	Mauchly
38	LOCATION	Switzerland
38	PERSON	Bonferroni
44	PERSON	Low
53	PERSON	Morbus Meulengracht
60	PERSON	Weathers
60	PERSON	T.M. Keane
60	PERSON	J.R. Davidson
60	PERSON	Blake
60	PERSON	Grof
60	PERSON	Ben Lomond
60	PERSON	Greer
60	PERSON	G.R.
60	PERSON	R. Tolbert
60	PERSON	Grof
60	LOCATION	Albany
60	PERSON	Metzner
60	PERSON	S. Adamson
60	PERSON	J. Holland
60	LOCATION	Rochester
60	PERSON	Mithoefer
60	PERSON	D. Hyman
60	PERSON	Berl
60	PERSON	Dumont
60	PERSON	Cami
60	PERSON	J Clin Psychopharmacol
60	PERSON	Dumont
60	PERSON	G.J.
60	PERSON	R.J. Verkes
60	PERSON	J Psychopharmacol
60	PERSON	C.M.
60	PERSON	Kirkpatrick
60	PERSON	Berl
60	PERSON	Kolbrich
60	PERSON	Liechti
60	PERSON	Mithoefer
60	LOCATION	M.C.
60	PERSON	J Psychopharmacol
60	PERSON	Tancer
60	PERSON	C.E. Johanson
60	PERSON	Berl
61	PERSON	Yubero-Lahoz
61	PERSON	Clin Pharmacokinet
61	PERSON	J Psychopharmacol
61	PERSON	Johanson
61	PERSON	K.P.
61	PERSON	J.G. Ramaekers
61	PERSON	Berl
61	PERSON	Grob
61	PERSON	Harris
61	PERSON	Berl
61	PERSON	Greer
61	PERSON	R. Tolbert
61	PERSON	Grob
61	LOCATION	C.S.
61	PERSON	Lester
61	PERSON	Ann Intern Med
61	PERSON	Peter Oehen
61	PERSON	Vollenweider
61	PERSON	Tancer
61	PERSON	C.E. Johanson
61	PERSON	Tancer
61	PERSON	C.E. Johanson
61	PERSON	Stolaroff
61	LOCATION	Sarasota
61	LOCATION	New York
61	PERSON	Blake
61	PERSON	Schnyder
61	PERSON	U.
61	PERSON	H. Moergeli
61	PERSON	Foa
61	LOCATION	E.B.
61	PERSON	Foa
61	LOCATION	E.B.
61	PERSON	Newman
61	PERSON	Brady
62	PERSON	Davidson
62	PERSON	Marshall
62	PERSON	R.D.
62	PERSON	Tucker
62	PERSON	Brunner
62	PERSON	S. Domhof
62	PERSON	F. Langer
62	LOCATION	New York
62	PERSON	Wiley
62	PERSON	Brunner
62	PERSON	U. Munzel
62	PERSON	Nichtparametrische Datenanalyse
62	LOCATION	Berlin
62	LOCATION	Los Angeles
62	PERSON	Marshall
62	PERSON	Weathers
62	LOCATION	Los Angeles
62	LOCATION	Krakow
62	PERSON	M. Hollifield
62	PERSON	Jama

—————— eof ———–