Direct Clinical Applications of Natural Language Processing in Common Neurological Disorders: Scoping Review (Preprint)

Опубликована: Окт. 2, 2023

Natural language processing (NLP), a branch of artificial intelligence that analyzes unstructured language, is being increasingly used in health care. However, the extent to which NLP has been formally studied neurological disorders remains unclear. We sought characterize studies applied diagnosis, prediction, or treatment common disorders. This review followed PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension Scoping Reviews) standards. The search was conducted using MEDLINE Embase on May 11, 2022. Studies use migraine, Parkinson disease, Alzheimer stroke transient ischemic attack, epilepsy, multiple sclerosis were included. excluded conference abstracts, papers, as well involving heterogeneous clinical populations indirect uses NLP. Study characteristics extracted analyzed descriptive statistics. did not aggregate measurements performance our due high variability study outcomes, main limitation study. In total, 916 identified, 41 (4.5%) met all eligibility criteria included final review. Of studies, most frequently represented attack (n=20, 49%), by epilepsy (n=10, 24%), disease (n=6, 15%), (n=5, 12%). found no migraine criteria. objective diagnosis phenotyping (n=17, 41%), prognostication (n=9, 22%), (n=4, 10%). 18 (44%) only machine learning approaches, 6 (15%) rule-based methods, 17 (41%) both. commonly implying potential role augmenting diagnostic accuracy settings with limited access expertise. also several gaps research, few addressing certain disorders, may suggest additional areas inquiry. Prospective Register (PROSPERO) CRD42021228703; https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=228703

Язык: Английский

Detecting multiple sclerosis disease activity and progression in progress notes from electronic medical records using natural language processing and machine learning

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2022, Номер unknown

Опубликована: Окт. 13, 2022

ABSTRACT Multiple sclerosis (MS) phenotypes provide useful disease descriptions but lack complete information regarding the continuing process. Disease activity and progression are meaningful modifiers of MS which can further guide prognosis, therapeutic decisions, clinical trial designs outcomes, were not explicitly documented in patients’ electronic medical records (EMRs). We aimed to detect patients with from notes EMR using Natural Language Processing Machine Learning models. Using randomly selected progress at University Rochester clinic, we integrated NLP machine learning technologies predict phenotype that represent progression. The method was evaluated by performance both models models, as well interpretability method. identified 460 287 adult patients. model had an average 0.92 precision, 0.87 recall, 0.89 F-score for entity extraction. It 0.85 0.84 relation sensitivities specificities classification algorithms predicting were: 67% 93% modifier “Active”, 61% 82% “Worsening”, 92% 98% “Progression”, 80% 94% “New MRI Lesion”, respectively. showed is capable detecting evidence notes. yielded interpretable largely clinically relevant features (symptoms conditions) persistently associated This holds promise facilitating screening participants potentially identifying early Author Summary disability be base impact outcomes. However, studies have shown neither nor their consistently record (EMR) chart often resides notes, requiring manual review experts increasing difficulty conducting research. In this paper, developed a generalized extraction, prediction pipeline, incorporating (NLP) shallow modifiers. Results demonstrated extracts progression, predicts satisfactory performance, encouraging portability interpretability. future, apply study high throughputs assessing modifying therapy utilization based on

Язык: Английский

Autoencoder Composite Scoring to Evaluate Prosthetic Performance in Individuals with Lower Limb Amputation

Bioengineering, Год журнала: 2022, Номер 9(10), С. 572 - 572

Опубликована: Окт. 18, 2022

We created an overall assessment metric using a deep learning autoencoder to directly compare clinical outcomes in comparison of lower limb amputees two different prosthetic devices—a mechanical knee and microprocessor-controlled knee. Eight were distilled into single seven-layer autoencoder, with the developed compared similar results from principal component analysis (PCA). The proposed methods used on data collected ten participants dysvascular transfemoral amputation recruited for prosthetics research study. This summary permitted cross-validated reconstruction all eight scores, accounting 83.29% variance. derived score is also linked functional ability this limited trial population, as improvements each base led increases metric. There was highly significant increase autoencoder-based when subjects (p < 0.001, repeated measures ANOVA). A traditional PCA interpretation but captured only 67.3% composite represents single-valued, succinct that can be useful holistic variable, individual scores datasets.

Язык: Английский