Page 28 - AIH-1-1

Artificial Intelligence in Health NLP in EHR

Table 2. (Continued)
Article Focus NLP level Limitations Future scope
[34] Development of an Semantic Negation detection, a German translation of Semantic interoperability
algorithm for information UMLS
extraction
[35] Identification of peripheral Semantic Captures only clinical visits Larger PAD cohort from multiple
arterial disease institutes
[37] Automatic extraction of Lexical and semantic Biases in the annotation process, small training Application of the method to
clinical findings from and test sets, and inclusion exclusively of chest various radiology reports other than
radiology reports CT reports chest CT involving structured and
unstructured clinical notes
[38] Early identification of Semantic Insufficient number of patients who had notes The approach can be applied
symptoms entered, patients with no follow-up records, to another disease using the
classifier built and tested locally, only focusing on classification method
a female patient, and inaccurate ICD coding
[39] Automatic identification of Semantic Replication without a positive set; complex Graph-based representation to
breast cancer and confusion in terminologies; no analysis of capture relation
pathology reports; and little semantic relations
[40] Identification of clinical Semantic Degree of uncertainty in diagnosis of pneumonia; Improved support for medical
assertion of pneumonia problems of false positive and negation decision-making
[41] Information extraction Semantic A small set of annotations, rate of false positives, Expand to all diagnostic imaging.
from radiology reports and ambiguities in natural language complicate
the negation detection task and subjectivity of
annotators
[42] NA ICU Domain knowledge, the hierarchical tree Cohort identification, automatic code
structure of medical codes assignment, and use of deep neural
network in the prediction model
[43] Detection of infection Semantic False positive occurrence, use of only radiology Conducting data analysis across
reports various report types to make
generalizations of the method, using
non-enriched text, using structured
or unstructured data
[45] Identification of respiratory Semantic Retrospective and single-site design NA
failure
[47] Development of Lexical and semantic Typing errors in findings Formalization of the syntactic
methodology for automatic structure of negated finding,
extraction enrichment of specialty dictionary,
and creation of more complete
dictionaries
[48] Extracting polyp Semantic The degree of accuracy depends on manual Applications of findings to other
information from data abstraction leads to errors and incorrect healthcare settings
colonoscopy reports assignments
[49] Identification of surgical site Lexical and semantic Use of clinical notes to study SSI, Various machine learning
infection relatively low F1 score approaches, sub-language supporting
techniques
[50] Extraction of fall Semantic Limited to one type of clinical note and one Investigation into the system and
information domain use of different machine learning
algorithms
[51] Surveillance of surgical site Semantic Annotation errors, mention level, and Removal of document and mention
infection document-level errors level errors
[53] Detection of hypoglycemia Semantic EMR from only one health system, lack of NA
diabetes duration
[54] Extract ion of clinical events Lexical, semantic, and NA Comparison with other language
morphological models
(Cont’d...)

Volume 1 Issue 1 (2024) 22 https://doi.org/10.36922/aih.2147