De-identification of patient notes with recurrent neural networks

Dernoncourt, Franck; Lee, Ji Young; Uzuner, Özlem; Szolovits, Peter

doi:10.1093/jamia/ocw156

Cited by 290 publications

(317 citation statements)

References 32 publications

Supporting

Mentioning

310

Contrasting

Order By: Relevance

“…[4,14] In the recent years, there has been an increase in the use of deep neural networks for a variety of NLP tasks, including NER. [5,6,7] Pre-trained word embeddings [8,9,15] have been used in traditional ML methods [16,17] and in neural networks, where Deconourt et al [18] has achieved better performance than previously published systems in de-identification of patient notes. Finally, Lee et al [19] have shown that featureaugmented neural networks can use a combination of word embeddings and domain-specific features to further improve the performance in de-identification of patient notes.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Recurrent neural networks with specialized word embeddings for health-domain named-entity recognition

Unanue

Borzeshi

Piccardi

2017

Journal of Biomedical Informatics

112

View full text Add to dashboard Cite

Background. Previous state-of-the-art systems on Drug Name Recognition (DNR) and Clinical Concept Extraction (CCE) have focused on a combination of text "feature engineering" and conventional machine learning algorithms such as conditional random fields and support vector machines. However, developing good features is inherently heavily time-consuming. Conversely, more modern machine learning approaches such as recurrent neural networks (RNNs) have proved capable of automatically learning effective features from either random assignments or automated word "embeddings". Objectives. (i) To create a highly accurate DNR and CCE system that avoids conventional, time-consuming feature engineering. (ii) To create richer, more specialized word embeddings by using health domain datasets such as MIMIC-III. (iii) To evaluate our systems over three contemporary datasets. Methods. Two deep learning methods, namely the Bidirectional LSTM and the Bidirectional LSTM-CRF, are evaluated. A CRF model is set as the baseline to compare the deep learning systems to a traditional machine learning approach. The same features are used for all the models. Results. We have obtained the best results with the Bidirectional LSTM-CRF model, which has outperformed all previously proposed systems. The specialized embeddings have helped to cover unusual words in DDI-DrugBank and DDI-MedLine, but not in the 2010 i2b2/VA IRB Revision dataset. Conclusion. We present a state-of-the-art system for DNR and CCE. Automated word embeddings has allowed us to avoid costly feature engineering and achieve higher accuracy. Nevertheless, the embeddings need to be retrained over datasets that are adequate for the domain, in order to adequately cover the domain-specific vocabulary.

show abstract

Section: Related Workmentioning

confidence: 99%

“…However, it has been proved that pre-trained word embeddings can improve the performance of the network. [5,12,18,19] In this section, we present the pre-trained embeddings employed in lieu of the random assignments.…”

Section: Word Featuresmentioning

confidence: 99%

Recurrent neural networks with specialized word embeddings for health-domain named-entity recognition

Unanue

Borzeshi

Piccardi

2017

Journal of Biomedical Informatics

112

View full text Add to dashboard Cite

show abstract

“…However, permissions to access other's data in a central warehouse are still cumbersome to obtain and de-identification efforts are either costly, error prone, or ineffective [6,7]. Human-based de-identification effort costs over 5,000 hours and US$500,000 on Medical Information Mart for Intensive Care III (MIMIC-III) dataset [8], which contains only about 50,000 patient visits and 100 million words [9]. and produces error recall ranging from 0.63 to 0.94 [10].…”

Section: Introductionmentioning

confidence: 99%

“…Machine-assisted de-identification shows varying results from time savings of 13.85-21.5% to results showing no improvement in either quality or time saved [11]. Machine learning algorithm-based automated deidentification can be very useful, but state-of-the-art deep learning-based de-identification models for unstructured data is still incapable of reaching to the level of privacy protection set by HIPAA safe harbor, which has roughly a 0.013% re-identification rate [9,12]. In the biomedical community, there is an urgent need for developing a new method to share information learned from local sources to generalize and scale up research effort.…”

Section: Introductionmentioning

confidence: 99%

Privacy-Preserving Predictive Modeling: Harmonization of Contextual Embeddings From Different Sources (Preprint)

Huang¹,

Lee²,

Wang³

et al. 2017

Preprint

View full text Add to dashboard Cite

Background: Data sharing has been a big challenge in biomedical informatics due to privacy concerns. Contextual embedding models have demonstrated very strong representative capability to describe medical concepts (and their context), and they have shown promise as an alternative way to support deep learning applications without the need to disclose original data. However, contextual embedding models acquired from individual hospitals cannot be directly combined because their embedding spaces are different and naive pooling renders combined embeddings useless.Objective: We present a novel approach to address these issues to promote sharing representation without sharing data. We can build a global model from representations learned from local private data without sacrificing privacy and synchronize information from multiple sources. Methods:We propose a methodology that harmonizes different local contextual embeddings into a global model. We use Word2Vec to generate contextual embeddings from each source and Procrustes to fuse different vector models into one common space by using a list of corresponding pairs as anchor points. With harmonized embeddings, we performed prediction analysis. Results:We used sequential medical events extracted from the Medical Information Mart for Intensive Care III database to evaluate the proposed methodology in predicting the next likely diagnosis of a new patient using either structured data or unstructured data. Under different experimental scenarios, we confirmed that the global model built from harmonized local models achieves more accurate prediction than local models and global model built from naive pooling.Conclusions: Such aggregation of local models using our unique harmonization can serve as the proxy for a global model, combining information from a wide range of institutions and information sources. It allows information unique to a certain hospital to become available to other sites, increasing the fluidity of information flow in health care.

show abstract

“…Recent research has been applying neural network methods to various clinical NLP tasks, such as named entity recognition (NER) [19,20], de-identification [21] and sentiment analysis [22][23][24]. Neural network methods are advantageous because they were shown to save significant time on task specific features extraction, achieved high performance, and are scalable to large applications [25].…”

Section: Introductionmentioning

confidence: 99%

Public Perception Analysis of Tweets During the 2015 Measles Outbreak: Comparative Study Using Convolutional Neural Network Models (Preprint)

Du¹,

Tang²,

Xiang³

et al. 2017

Preprint

View full text Add to dashboard Cite

Background: Timely understanding of public perceptions allows public health agencies to provide up-to-date responses to health crises such as infectious diseases outbreaks.Social media such as Twitter provides an unprecedented way for the prompt assessment of the large-scale public response.Objective: This study has two aims: 1) to develop a scheme for a comprehensive public perception analysis of a measles outbreak from Twitter data; 2) to demonstrate the superiority of the neural network-based approach (compared with conventional machinelearning methods) on measles outbreak-related tweets classification tasks with a relatively small and highly unbalanced gold standard training set. Methods Conclusions:Compared to conventional machine-learning methods, our CNN models showed superiority on measles related tweet classification tasks with a relatively small and highly unbalanced gold standard. Our proposed scheme and neural network based tweets classification system could be used for the analysis of tweets about other infectious diseases such as influenza and Ebola.

show abstract

De-identification of patient notes with recurrent neural networks

Abstract: Our findings support the use of ANNs for de-identification of patient notes, as they show better performance than previously published systems while requiring no manual feature engineering.

Cited by 290 publications

References 32 publications

Recurrent neural networks with specialized word embeddings for health-domain named-entity recognition

Recurrent neural networks with specialized word embeddings for health-domain named-entity recognition

Privacy-Preserving Predictive Modeling: Harmonization of Contextual Embeddings From Different Sources (Preprint)

Public Perception Analysis of Tweets During the 2015 Measles Outbreak: Comparative Study Using Convolutional Neural Network Models (Preprint)

Contact Info

Product

Resources

About