![]() Natural language processing (NLP) refers to a set of techniques for preparing text data and converting it into a structured form suitable for subsequent machine learning. Clinical decision support (CDS) systems are not yet commonly used in radiology, and the unstructured form of electronic radiology referrals requires preprocessing. The information contained in radiology referrals is manually used during justification, for assessing their compliance with clinical imaging guidelines which aim to improve patient care through evidence-based recommendations of radiology resources. Avoiding unnecessary exposure to ionising radiation is the primary evidence-based intervention that reduces cancer risk. To improve justification of CT examinations in the European Union, the European Commission recently funded a 3-year project to co-ordinate audits of justification of CT examinations and to develop a common methodology for auditing justification of CT referrals. Furthermore, individual local audits within Europe indicate poor justification practice with the rate of unjustified CT examinations between 7 and 30%. National audits from Northern Ireland, Sweden, and Luxembourg, report 6%, 19%, and 39% of unjustified CT examinations, respectively. Increasing CT frequency poses additional population dose burden, as CT is the largest dose contributor. A similar trend is seen in the UK where the CT scan frequency between 20 increased by approximately 74%, while a 20% increase occurred in the USA in the 2006–2016 period. Since 2009, the number of CT examinations carried out in Ireland has almost doubled. Many patients undergo multiple CT scans therefore, their cumulative risk of developing a radiation-induced cancer is significantly higher. This offers potential for automated justification analysis of CT referrals in clinical departments.Ĭomputed tomography (CT) scans are associated with relatively high radiation doses, and as a result, patients are potentially at greater lifetime risk of developing a radiation-induced cancer. Traditional ML models can accurately predict justification of unstructured brain CT referrals. The BOW + CSW + SC + SVM outperformed other binary models with a weighted accuracy of 92%, a sensitivity of 91%, a specificity of 93%, and an AUC of 0.948. ![]() The agreement between the annotators was strong ( κ = 0.835). In total, 253 (67.5%) examinations were deemed justified, 75 (20.0%) as unjustified, and 47 (12.5%) as maybe justified. A test set (300/75) was used to compute weighted accuracy, sensitivity, specificity, and the area under the curve (AUC). ![]() Logistic regression, random forest, and support vector machines (SVM) were used to predict the justification of referrals. ![]() Text preprocessing techniques, including custom stop words (CSW) and spell correction (SC), were applied to the referral text. Referrals were represented as bag-of-words (BOW) and term frequency-inverse document frequency models. Cohen’s kappa was computed to measure inter-rater reliability. Two human experts retrospectively analysed justification of 375 adult brain CT referrals performed in a tertiary referral hospital during the 2019 calendar year, using a cloud-based platform for structured referring. We aimed to retrospectively audit justification of brain CT referrals by applying natural language processing and traditional machine learning (ML) techniques to predict their justification based on the audit outcomes. Manual justification audits of radiology referrals are time-consuming and require financial resources. With a significant increase in utilisation of computed tomography (CT), inappropriate imaging is a significant concern. ![]()
0 Comments
Leave a Reply. |