Natural Language Processing in Radiology: A Systematic Review

Pons, Ewoud; Braun, Loes; Hunink, M. G. Myriam; Kors, Jan A.

doi:10.1148/radiol.16142770

Cited by 467 publications

(343 citation statements)

References 61 publications

Supporting

Mentioning

336

Contrasting

Unclassified

Order By: Relevance

“…For example, incidental imaging findings encountered in the emergency department setting are at risk for delayed or failed follow-up for a variety of reasons, some of which include discontinuity of care and limited availability of diagnostic reports and patient records [11].…”

Section: Communication Errorsmentioning

confidence: 99%

The role of radiology in diagnostic error: a medical malpractice claims review

Siegal¹,

Stratchko

DeRoo

2017

Diagnosis

View full text Add to dashboard Cite

Abstract:Just as radiologic studies allow us to see past the surface to the vulnerable and broken parts of the human body, medical malpractice claims help us see past the surface of medical errors to the deeper vulnerabilities and potentially broken aspects of our healthcare delivery system. And just as the insights we gain through radiologic studies provide focus for a treatment plan for healing, so too can the analysis of malpractice claims provide insights to improve the delivery of safe patient care. We review 1325 coded claims where Radiology was the primary service provider to better understand the problems leading to patient harm, and the opportunities most likely to improve diagnostic care in the future.Keywords: diagnostic error; malpractice claims; radiology. OverviewThe medical diagnostic process involves a complex network of interactions between the patient and the healthcare system. This process is also dynamic, requiring one or more cycles of patient interaction, informationgathering and data synthesis in order to understand the intricacies of each patient's clinical picture and pathology. Failures can occur at any point along the continuum of care, each of which has the potential to result in inaccurate or delayed diagnosis as well as inappropriate treatment. While radiology typically does not play the initial role in the diagnostic process, misinterpretation or delayed communication of imaging findings can certainly lead to a breakdown in the progression towards clarity of diagnosis and appropriate patient care.Analysis of the CRICO Comparative Benchmarking System (CBS) determined that 29,777 medical malpractice cases, asserted between 2010 and 2014, had completed an in-depth review by CRICO's team of Clinical Taxonomy Specialists. Reviewing the medical and legal files of each of these cases, an experienced clinician used CRICO's propriety coding taxonomy to capture and code multiple case attributes including allegation, patient demographics, diagnosis and injury, location, tests and services, and the key causation factors contributing to the clinical error or failure.Of the 29,777 medical malpractice cases available for analysis, 1325 cases named Radiology as the Primary Responsible Service -42% resulted in high severity (based on National Associationn of Insurance Commissioners clinical injury severity score) clinical injuries including 235 deaths (Figure 1). Diagnostic related events represent nearly 60% of the 1325 radiology claims, followed by procedural issues (22%), equipment issues, (7%) and falls and safety issues (6%). In those cases involving diagnostic radiology, nearly 50% of the cases involved one of these four modalities: computed tomography (CT) scans (20%), mammography (11%), magnetic resonance imaging (MRI) (10%) and diagnostic ultrasound (4%). Cases occurred in a variety of settings though ambulatory cases were the most common at 63% followed by inpatient (26%) and emergency department (11%).In many cases, Radiology is not the only clinical service identified as "responsible" o...

show abstract

Section: Communication Errorsmentioning

confidence: 99%

The role of radiology in diagnostic error: a medical malpractice claims review

Siegal¹,

Stratchko

DeRoo

2017

Diagnosis

View full text Add to dashboard Cite

show abstract

“…Natural language processing methods can be used to find quality indicators of radiological practice (23), automatically analyze content (24), classify free-text reports (25) or assess the presence of sufficient clinical recommendations in radiological reports (26). Such methods successfully use more complex language models and retrieve contextual information, but they need to be tailored for specific conditions and require dedicated sets of examples for training and optimization (23).…”

Section: (Css) Distributions Of Conventional Free Text Reports (Cftr)mentioning

confidence: 99%

“…In contrast, the query-based retrieval approach can be more flexible, with fast, expert-driven customization, while being more suitable for smaller datasets (23). Therefore, we used the latter approach that uses key terms taken from guidelines to build a query-based quality measure.…”

Section: (Css) Distributions Of Conventional Free Text Reports (Cftr)mentioning

confidence: 99%

Objective Comparison Using Guideline-based Query of Conventional Radiological Reports and Structured Reports

Maros

Wenz

Förster

et al. 2018

In Vivo

View full text Add to dashboard Cite

Abstract. Background: This feasibility study of text-miningbased scoring algorithm provides an objective comparison of structured reports (SR) and conventional free-text reports (cFTR) by means of guidelineStructured reporting (SR) represents a new direction in communicating radiological reports to clinicians in a way that is clear and uniform (1). However, follow-up studies on SR have shown mixed results regarding adaptation and adherence (2-4). In particular, neuroradiological residents most commonly complain that SRs are overly constraining and time-consuming (5). In spite of that, there is an increasing tendency to use online solutions to generate SR templates (6). For instance, "CT brain" and "MR brain" were the third and fifth most frequently viewed SR templates in the Radiological Society of North America (RSNA) online library (7). SR templates can serve as core frameworks, but because of heterogeneous representations of diseases and co-morbidities, there is a substantial need to allow for customization of each report -primarily in the form of additional free text. Consequently, the distinction between conventional free-text (narrative) reports (cFTR) and SR becomes even more blurred, which makes their objective comparison even harder. This raises two further questions: Are SRs indeed 'better' than cFTRs? How can cFTRs be compared with SRs in an objective, fast and scalable way if SRs are often a mixture of structured and free (narrative) text?Firstly, there is a need to define how to objectively quantify 'better'. Commonly the method of choice for such quality assessments is either an expert opinion-based rating (5, 8) or a survey-based evaluation by physicians (4). Although both methods provide valuable insight, they are very laborious and time-consuming for experts and do not scale well for large datasets (4,8). Therefore, we aimed to implement an approach using the adherence to imaging and clinical guidelines as a quality measure.Secondly, in order to objectively compare cFTR with SR, we suggest an evaluation based on widely used text-mining technique using the 'bag of words' representation and cosine similarity (9, 10). The cosine similarity provides the core of many information retrieval systems (11,12). Reports can be 843 This article is freely accessible online.

show abstract

“…By the implementation of the structured template, clinical data in EMR can be standardized to resolve ambiguity, provide semantic interoperability, and prevent data entry errors. Second, to extract the meaningful information from the existing text documents, clinical natural language processing (NLP) methods should be developed [20][21][22], or the simple regular expression can also be applied [23]. In Korea, the simple regular expression can be more practical and promising at the current stage due to the lack of Korean NLP research.…”

Section: Clinical Datamentioning

confidence: 99%

Status and Direction of Healthcare Data in Korea for Artificial Intelligence

Park

Shin

2017

Hanyang Med Rev

View full text Add to dashboard Cite

Predicting temporal patterns across various domains poses significant challenges due to their nuanced and often nonlinear trajectories. To address this challenge, prediction frameworks have been continuously refined, employing data-driven statistical methods, mathematical models, and machine learning. Recently, as one of the challenging systems, shared transport systems such as public bicycles have gained prominence due to urban constraints and environmental concerns. Predicting rental and return patterns at bicycle stations remains a formidable task due to the system's openness and imbalanced usage patterns across stations. In this study, we propose a deep learning framework to predict rental and return patterns by leveraging cartogram approaches. The cartogram approach facilitates the prediction of demand for newly installed stations with no training data as well as long-period prediction, which has not been achieved before. We apply this method to public bicycle rentaland-return data in Seoul, South Korea, employing a spatial-temporal convolutional graph attention network. Our improved architecture incorporates batch attention and modified node feature updates for better prediction accuracy across different time scales. We demonstrate the effectiveness of our framework in predicting temporal patterns and its potential applications.

show abstract

Natural Language Processing in Radiology: A Systematic Review

Cited by 467 publications

References 61 publications

The role of radiology in diagnostic error: a medical malpractice claims review

The role of radiology in diagnostic error: a medical malpractice claims review

Objective Comparison Using Guideline-based Query of Conventional Radiological Reports and Structured Reports

Status and Direction of Healthcare Data in Korea for Artificial Intelligence

Contact Info

Product

Resources

About