< Back to previous page

Publication

Towards standardised radiologic evaluation and reporting of acute traumatic brain injuries

Book - Dissertation

Neuroimaging techniques play a pivotal role in the diagnosis and follow-up of patients with Traumatic Brain Injury (TBI). Imaging findings determine patient management and influence the clinical course. Unfortunately, conventional evaluation of imaging studies and free-text radiologic reporting is problematic in patients with TBI, mainly due to the significant differences in content, length, style and language of the generated radiology reports. Moreover, there is substantial inter-observer variation in how pathology is characterised, even between experts. In order to advance clinical decision-making and research into TBI, a more complete, precise and consistent characterisation of brain pathology was urgently needed. In 2010, a multidisciplinary task force, under the aegis of the National Institute of Health (NIH) and the National Institute of Neurological Disorders and Stroke (NINDS), created a framework for a more structured way of radiologic reporting in clinical trials, by providing a standardised language of "common data elements" (CDEs). Despite the many presumed benefits of this standardised framework, a thorough investigation and validation was still lacking. We performed a standardised central radiology review, based on the NIH/NINDS CDEs, on over 4,000 acute non-contrast computed tomography (NCCT) scans, uploaded to a central imaging database for the large pan-European CENTER-TBI study. We explored inter- and intra-observer agreement, compared centralised versus local assessment, and we created regularised logistic regression models to investigate the prognostic relevance of the different NIH/NINDS CDEs. In addition, we also developed and tested machine learning algorithms, intended to objectively quantify relevant imaging characteristics. Our results indicate that on-site radiologic evaluation and reporting suffers from substantial inconsistencies, which is highly detrimental in multi-centre studies. Conversely, we showed that an independent central radiology review process, using NIH/NINDS CDEs, offers a more consistent way of characterising pathology, with high levels of inter-and intra-observer agreement. We also demonstrated that, on a large scale, this kind of characterisation allows for extensive data mining and the development of strong clinical predictive models. Furthermore, our automated machine learning-based approach achieved good performance for the segmentation of lesion volumes, cisternal volumes, and midline shift measurement. In conclusion, our work shows that standardised radiologic evaluation and reporting, using the NIH/NINDS CDEs, combined with automation of certain aspects of radiologic evaluation, is very promising and can significantly advance clinical research. Based on our findings, we made recommendations for clinical radiology reporting, and developed an image interpretation checklist, accompanied by a comprehensive and fully illustrated patho-anatomic atlas, with over 200 images from the CENTER-TBI study. Our efforts can help physicians and researchers to reliably and reproducibly interpret NCCT scans of brain-injured patients, thus paving the way towards a more globally accepted standardised evaluation and reporting of acute traumatic brain injuries.
Number of pages: 202
ISBN:978-94-6400-755-8
Publication year:2020
Keywords:Doctoral thesis
Accessibility:Open