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

Abstract

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.

Survey of Post-OCR Processing Approaches

“…A character-level bidirectional LSTM language model is developed and applied to post-process digitised French clinical texts by D'hondt et al [39,40]. Given clean texts (i.e., digital collections containing no errors), they automatically create the training material by randomly applying edit operations (i.e., deletion, insertion, substitution).…”

“…Some techniques are proposed to generate artificial material such as randomly deleting, inserting, and substituting characters from a given word [39][40][41]; mimicking realistic errors from repetition texts: picking up an alternative from the list of frequent replaceable characters for a given character [57]; or using reversed error model with the input being GT word ngrams [52].…”

Named Entity Recognition for Chinese biomedical patents

“…In the general domain, there are several well developed tools that can detect and correct possible OCR errors automatically. But for domain-specific text, popular existing tools based on simple rules can not handle OCR errors well, especially if there is complex and difficult terminology use (D'hondt et al, 2016;Thompson et al, 2015). Facing this problem, it might be necessary to build an OCR correction tool adapted for this domain and genre specifically (Zhang et al, 2019).…”

Analysis of Context-Dependent Errors in the Medical Domain in Spanish: A Corpus-Based Study

“…Various investigations have been carried out on automatic detection and correction in clinical documentation and, specifically, in clinical reports. Most of these studies have been developed with corpora in English (e.g., Fivez et al, 2017;Workman et al, 2019), but these issues have been also explored for French (D'Hondt et al, 2016), Russian (Balabaeva et al, 2020), Swedish (Dziadek et al, 2017), Dutch (Fivez et al, 2017), Hungarian (Siklo´si et al, 2016), and Persian (Yazdani et al, 2020). These works highlight the substantial number of linguistic errors that the corpora comprising clinical reports usually contain, with studies whose error rate is around 5% (Lai et al, 2015) or even 10% (Ruch et al, 2003).…”