Observer Performance in Computed Tomography Head Reporting

Abstract

Aim: To audit the reporting results of a cohort of radiographers (n=6) completing an accredited academic programme in clinical reporting of Computed Tomography (CT) head examinations.Methods: An audit of retrospective academic image case banks and prospective random clinical workload case banks. Both the academic test banks and clinical workload banks included a wide range of normal and abnormal cases of different levels of difficulty and pathology. Abnormalities included: haemorrhage, fractures, lesions, infarctions, degeneration, and normal variants from a variety of referral sources. True positive and negative, as well as false positive and negative fractions were used to mark the reports, which were analysed for accuracy against a reference standard. Further interobserver variability was assessed using Cohens Kappa, one-way analysis of variance and Tukey for multiple comparisons and significance testing at 95% confidence intervals. Results:The mean accuracy score for all radiographers (n=6) and reports (n=3,008) was 90.7% (95%CI 88.3%-93.0%). Mean sensitivity and specificity rate was 86.9% (95%CI 85.8%-88.2%), and 94% (95%CI 89.6%-98.3%) respectively. The most common errors were associated with herniation, lacunar infarctions and subtle fractures (false negatives) and involutional changes, subtle infarctions, and ventricular dilation (false positives). Conclusions:The results suggest appropriately trained radiographers can successfully undertake to report CT head examinations to a high standard. The adoption of both academic and clinical workload image banks that reflect disease examples and the prevalence that may logically be encountered in practice offers the potential for an accurate measure of performance of radiographer s abilities.

Object recognition ability predicts category learning with medical images

“…Practitioners may even disagree with their own initial judgement when asked to review images a second time (Abujudeh et al, 2010 ). Although precise estimates of the prevalence of medical imaging errors are difficult to obtain, as errors vary widely based on test, practice setting, and population, estimates of real-world error rates range from < 1% to around 10% (Gergenti & Olympia, 2019 ; Lamoureux et al, 2021 ; Lockwood, 2017 ). Error rates can be higher still when the relevant disease is rare in the studied population (Kolb et al, 2002 ).…”