Language Modelling for the Needs of OCR of Medical Texts

Piasecki, Maciej; Godlewski, Grzegorz

doi:10.1007/11946465_25

Cited by 4 publications

(5 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To the best of our knowledge, the only existing OCR error detection and correction systems for medical texts focus on either OCR correction for historical text with adapted language models (Thompson et al, 2015) or OCR recognition of handwritten notes by doctors, which is not surprising given the absence of large OCRed text corpora in this domain. Notable work in this area was carried out by Piasecki et al (2006) who examined the construction of word-level language models to improve OCR correction of Polish handwritten medical notes. They found that the repetitive character sequences and recurrent structure of medical notes greatly aided the construction of language models but that this positive effect is domain-specific and does not carry over the similar corpora in a different medical subdomain.…”

Section: Introductionmentioning

confidence: 99%

“…In the first step, a potential OCR error is detected using either a lookup in a domain-specific lexicon (Kissos and Dershowitz, 2016) or unigram language model (Bassil and Alwani, 2012), and/or by consulting information from the OCR process, i.e., the confidence scores of the recognized characters. The second step, candidate generation, also heavily depends on external resources, either by generating potential candidate replacements for the erroneous words from a lexicon (Piasecki and Godlewski, 2006) or by learning and using a mapping of characters that were often interchanged during the OCR process to generate potential candidates with string distance metrics (Kukich, 1992). Such mappings are known as 'character confusions' but need to be learned over a training corpus of a considerable size before they can become effective (Evershed and Fitch, 2014).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Low-resource OCR error detection and correction in French Clinical Texts

D’Hondt¹,

Grouin²,

Grau³

2016

Proceedings of the Seventh International Workshop on Health Text Mining and Information Analysis

View full text Add to dashboard Cite

In this paper we present a simple yet effective approach to automatic OCR error detection and correction on a corpus of French clinical reports of variable OCR quality within the domain of foetopathology. While traditional OCR error detection and correction systems rely heavily on external information such as domain-specific lexicons, OCR process information or manually corrected training material, these are not always available given the constraints placed on using medical corpora. We therefore propose a novel method that only needs a representative corpus of acceptable OCR quality in order to train models. Our method uses recurrent neural networks (RNNs) to model sequential information on character level for a given medical text corpus. By inserting noise during the training process we can simultaneously learn the underlying (character-level) language model and as well as learning to detect and eliminate random noise from the textual input. The resulting models are robust to the variability of OCR quality but do not require additional, external information such as lexicons. We compare two different ways of injecting noise into the training process and evaluate our models on a manually corrected data set. We find that the best performing system achieves a 73% accuracy.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Low-resource OCR error detection and correction in French Clinical Texts

D’Hondt¹,

Grouin²,

Grau³

2016

Proceedings of the Seventh International Workshop on Health Text Mining and Information Analysis

View full text Add to dashboard Cite

show abstract

“…Notable work in this area was carried out by Piasecki et al (2006) who examined the construction of word-level language models to improve OCR correction of Polish handwritten medical notes. They 62 found that the repetitive character sequences and recurrent structure of medical notes greatly aided the construction of language models but that this positive effect is domain-specific and does not carry over the similar corpora in a different medical subdomain.…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Proceedings of the Seventh International Workshop on Health Text Mining and Information Analysis

2016

View full text Add to dashboard Cite

an interdisciplinary forum for researchers interested in automated processing of health documents. Health documents encompass electronic health records, clinical guidelines, spontaneous reports for pharmacovigilance, biomedical literature, health forums/blogs or any other type of health-related documents. The LOUHI workshop series fosters interactions between the Computational Linguistics, Medical Informatics and Artificial Intelligence communities. Following the six previous edition of the workshop which were co-located with SMBM 2008 in Turku, Finland, with NAACL 2010 in Los Angeles, California, with Artificial Intelligence in Medicine (AIME 2011) in Bled, Slovenia, during NICTA Techfest 2013 in Sydney, Australia, co-located with EACL 2014 in Gothenburg, Sweden, and with EMNLP 2015 in Lisbon, Portugal, this workshop is co-located this year with EMNLP 2016 in Austin, Texas.The aim of the LOUHI 2016 workshop is to bring together research work on topics related to health documents, particularly emphasizing multidisciplinary aspects of health documentation and the interplay between nursing and medical sciences, information systems, computational linguistics and computer science. The topics include, but are not limited to, the following Natural Language Processing techniques and related areas:• Techniques supporting information extraction, e.g. named entity recognition, negation and uncertainty detection• Classification and text mining applications (e.g. diagnostic classifications such as ICD-10 and nursing intensity scores) and problems (e.g. handling of unbalanced data sets)• Text representation, including dealing with data sparsity and dimensionality issues• Domain adaptation, e.g. adaptation of standard NLP tools (incl. tokenizers, PoS-taggers, etc) to the medical domain• Information fusion, i.e. integrating data from various sources, e.g. structured and narrative documentation• Unsupervised methods, including distributional semantics• Evaluation, gold/reference standard construction and annotation• Syntactic, semantic and pragmatic analysis of health documents• Anonymization/de-identification of health records and ethics• Supporting the development of medical terminologies and ontologies• Individualization of content, consumer health vocabularies, summarization and simplification of text• NLP for supporting documentation and decision making practices iii• Predictive modeling of adverse events, e.g. adverse drug events and hospital acquired infectionsThe call for papers encouraged authors to submit papers describing substantial and completed work but also focus on a contribution, a negative result, a software package or work in progress. We also encouraged to report work on low-resourced languages, addressing the challenges of data sparsity and language characteristic diversity.We received 21 submissions. Each submission went through a double-blind review process which involved three program committee members. Based on comments and rankings supplied by the reviewers, we accepted 13 papers. The overall acceptanc...

show abstract

“…Due to Zipf's Law, the sparse data problem is rarely avoided, while this problem brings great difficulties in improving the performance of the disambiguation and OOV identification. Ontology [14], as a concept modeling tool that can present information systems on the semantic and knowledge level [15][16][17], captures the attention of many researchers and play an import role in Knowledge Engineering, Digital Libraries, Software Reuse, Information Retrieval (IR), Semantic Web, the Interoperability of Heterogeneous Information and so on.…”

Section: Introductionmentioning

confidence: 99%

A New Field Word Segmentation Model Based on Ontology in Digital Library

Wang¹

2012

IJACT

View full text Add to dashboard Cite

This paper studied a novel method based on ontology and syntax analysis. According to the disadvantages of the method based on character string matching, the method based on comprehension and the method based on statistic, we not only gave a formal definition of the model and the functions on it, but also gave a semantic query algorithm based on ontology. With the introduction of ontology technology Chinese word segmentation and innovative mode, the experimental results indicate that the method largely eliminates semantic ambiguity, avoids losing problem due to ignore the environment information in traditional Chinese automatic segmentation method.

show abstract

Language Modelling for the Needs of OCR of Medical Texts

Cited by 4 publications

References 5 publications

Low-resource OCR error detection and correction in French Clinical Texts

Low-resource OCR error detection and correction in French Clinical Texts

Proceedings of the Seventh International Workshop on Health Text Mining and Information Analysis

A New Field Word Segmentation Model Based on Ontology in Digital Library

Contact Info

Product

Resources

About