Abstract Background Diabetic retinopathy (DR) is a leading cause of blindness in American adults. If detected, DR can be treated to prevent further damage causing blindness. There is an increasing interest in developing artificial intelligence (AI) technologies to help detect DR using electronic health records. The lesion-related information documented in fundus image reports is a valuable resource that could help diagnoses of DR in clinical decision support systems. However, most studies for AI-based DR diagnoses are mainly based on medical images; there is limited studies to explore the lesion-related information captured in the free text image reports. Methods In this study, we examined two state-of-the-art transformer-based natural language processing (NLP) models, including BERT and RoBERTa, compared them with a recurrent neural network implemented using Long short-term memory (LSTM) to extract DR-related concepts from clinical narratives. We identified four different categories of DR-related clinical concepts including lesions, eye parts, laterality, and severity, developed annotation guidelines, annotated a DR-corpus of 536 image reports, and developed transformer-based NLP models for clinical concept extraction and relation extraction. We also examined the relation extraction under two settings including ‘gold-standard’ setting—where gold-standard concepts were used–and end-to-end setting. Results For concept extraction, the BERT model pretrained with the MIMIC III dataset achieve the best performance (0.9503 and 0.9645 for strict/lenient evaluation). For relation extraction, BERT model pretrained using general English text achieved the best strict/lenient F1-score of 0.9316. The end-to-end system, BERT_general_e2e, achieved the best strict/lenient F1-score of 0.8578 and 0.8881, respectively. Another end-to-end system based on the RoBERTa architecture, RoBERTa_general_e2e, also achieved the same performance as BERT_general_e2e in strict scores. Conclusions This study demonstrated the efficiency of transformer-based NLP models for clinical concept extraction and relation extraction. Our results show that it’s necessary to pretrain transformer models using clinical text to optimize the performance for clinical concept extraction. Whereas, for relation extraction, transformers pretrained using general English text perform better.
more »
« less
Document-Level Event Argument Extraction by Conditional Generation
Event extraction has long been treated as a
sentence-level task in the IE community. We
argue that this setting does not match human
information seeking behavior and leads
to incomplete and uninformative extraction
results. We propose a document-level neural
event argument extraction model by formulating
the task as conditional generation
following event templates. We also compile
a new document-level event extraction
benchmark dataset WIKIEVENTS which includes
complete event and coreference annotation.
On the task of argument extraction, we
achieve an absolute gain of 7.6% F1 and 5.7%
F1 over the next best model on the RAMS
and WIKIEVENTS datasets respectively. On
the more challenging task of informative argument
extraction, which requires implicit coreference reasoning, we achieve a 9.3% F1 gain over the best baseline. To demonstrate the portability of our model, we also create the first end-to-end zero-shot event extraction framework and achieve 97% of fully supervised model’s trigger extraction performance and 82% of the argument extraction performance given only access to 10 out of the 33 types on ACE.
more »
« less
- NSF-PAR ID:
- 10311076
- Date Published:
- Journal Name:
- Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, {NAACL-HLT} 2021
- Volume:
- 2021
- Issue:
- 1
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)Abstract Objective Cancer is a leading cause of death, but much of the diagnostic information is stored as unstructured data in pathology reports. We aim to improve uncertainty estimates of machine learning-based pathology parsers and evaluate performance in low data settings. Materials and methods Our data comes from the Urologic Outcomes Database at UCSF which includes 3232 annotated prostate cancer pathology reports from 2001 to 2018. We approach 17 separate information extraction tasks, involving a wide range of pathologic features. To handle the diverse range of fields, we required 2 statistical models, a document classification method for pathologic features with a small set of possible values and a token extraction method for pathologic features with a large set of values. For each model, we used isotonic calibration to improve the model’s estimates of its likelihood of being correct. Results Our best document classifier method, a convolutional neural network, achieves a weighted F1 score of 0.97 averaged over 12 fields and our best extraction method achieves an accuracy of 0.93 averaged over 5 fields. The performance saturates as a function of dataset size with as few as 128 data points. Furthermore, while our document classifier methods have reliable uncertainty estimates, our extraction-based methods do not, but after isotonic calibration, expected calibration error drops to below 0.03 for all extraction fields. Conclusions We find that when applying machine learning to pathology parsing, large datasets may not always be needed, and that calibration methods can improve the reliability of uncertainty estimates.more » « less
-
Sentence-level relation extraction (RE) aims at identifying the relationship between two entities in a sentence. Many efforts have been devoted to this problem, while the best performing methods are still far from perfect. In this paper, we revisit two problems that affect the performance of existing RE models, namely entity representation and noisy or ill-defined labels. Our improved RE baseline, incorporated with entity representations with typed markers, achieves an F1 of 74.6% on TACRED, significantly outperforms previous SOTA methods. Furthermore, the presented new baseline achieves an F1 of 91.1% on the refined Re-TACRED dataset, demonstrating that the pretrained language models (PLMs) achieve high performance on this task. We release our code to the community for future research.more » « less
-
null (Ed.)Determining whether an event in a news article is a foreground or background event would be useful in many natural language processing tasks, for example, temporal relation extraction, summarization, or storyline generation. We introduce the task of distinguishing between foreground and background events in news articles as well as identifying the general temporal position of background events relative to the foreground period (past, present, future, and their combinations). We achieve good performance (0.73 F1 for background vs. foreground and temporal position, and 0.79 F1 for background vs. foreground only) on a dataset of news articles by leveraging discourse information in a featurized model. We release our implementation and annotated data for other researchersmore » « less
-
more » « less
Abstract Motivation Best performing named entity recognition (NER) methods for biomedical literature are based on hand-crafted features or task-specific rules, which are costly to produce and difficult to generalize to other corpora. End-to-end neural networks achieve state-of-the-art performance without hand-crafted features and task-specific knowledge in non-biomedical NER tasks. However, in the biomedical domain, using the same architecture does not yield competitive performance compared with conventional machine learning models.
Results We propose a novel end-to-end deep learning approach for biomedical NER tasks that leverages the local contexts based on n-gram character and word embeddings via Convolutional Neural Network (CNN). We call this approach GRAM-CNN. To automatically label a word, this method uses the local information around a word. Therefore, the GRAM-CNN method does not require any specific knowledge or feature engineering and can be theoretically applied to a wide range of existing NER problems. The GRAM-CNN approach was evaluated on three well-known biomedical datasets containing different BioNER entities. It obtained an F1-score of 87.26% on the Biocreative II dataset, 87.26% on the NCBI dataset and 72.57% on the JNLPBA dataset. Those results put GRAM-CNN in the lead of the biological NER methods. To the best of our knowledge, we are the first to apply CNN based structures to BioNER problems.
Availability and implementation The GRAM-CNN source code, datasets and pre-trained model are available online at: https://github.com/valdersoul/GRAM-CNN.
Supplementary information Supplementary data are available at Bioinformatics online.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.18653/v1/2021.naacl-main.69
