Abstract
Radiology reports contain important clinical information about patients which are often tied through spatial expressions. Spatial expressions (or triggers) are mainly used to describe the positioning of radiographic findings or medical devices with respect to some anatomical structures. As the expressions result from the mental visualization of the radiologist's interpretations, they are varied and complex. The focus of this work is to automatically identify the spatial expression terms from three different radiology sub-domains. We propose a hybrid deep learning-based NLP method that includes - 1) generating a set of candidate spatial triggers by exact match with the known trigger terms from the training data, 2) applying domain-specific constraints to filter the candidate triggers, and 3) utilizing a BERT-based classifier to predict whether a candidate trigger is a true spatial trigger or not. The results are promising, with an improvement of 24 points in the average F1 measure compared to a standard BERT-based sequence labeler.
Highlights
Radiology reports contain a radiologist’s interpretations of an imaging study of a patient
We focus on extracting varied spatial expressions from radiology reports using a sequence labeling method as well as a hybrid approach that first applies domain-specific rules to extract the candidate triggers and later employs a deep learning-based classifier to judge every candidate
This work proposes a BERT-based hybrid method to extract spatial expressions from radiology reports. This method achieves satisfactory performance with an average F1 measure of 81.10 over 10-fold cross validation (CV)
Summary
Radiology reports contain a radiologist’s interpretations of an imaging study of a patient. The radiological entities whose positions are described mainly include radiographic findings (e.g., clinical findings like interstitial emphysema and imaging observations like groundglass opacity) and medical devices (e.g., endotracheal tube and central venous catheter). The focus of this work is to investigate different automatic approaches to extract the spatial expressions from the report sentences along with highlighting the various challenges involved in this task. These extracted spatial expressions, if predicted accurately, can facilitate clinical applications such as automatic labeling of radiographic images for training image classifiers (Wang et al, 2017)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have