8th Baltic Congress of Radiology

08.10.2022  Tallinn, Estonia

Presentation

Terrace Hall

Synopsis: The presentation will introduce the study that aimed to build a deep learning model to derive labels from neuroradiology reports and assign these to the corresponding examinations, overcoming a bottleneck to computer vision model development. Reference-standard labels were generated by a team of neuroradiologists for model training and evaluation. Three thousand examinations were labelled for the presence or absence of any abnormality by manually scrutinising the corresponding radiology reports (‘reference-standard report labels’). Separately, 2000 reports were labelled for the presence or absence of 7 specialised categories of abnormality (acute stroke, mass, atrophy, vascular abnormality, small vessel disease, white matter inflammation, encephalomalacia), with a subset of these examinations (n = 700) also assigned reference-standard image labels. The accurate classification was achieved for all categories when tested against reference-standard report labels. A minimal drop in performance was seen for three categories (atrophy, encephalomalacia, vascular) when tested against reference-standard image labels, highlighting discrepancies in the original reports. Once trained, the model assigned labels to 121,556 examinations in under 30 min. In conclusion, the model created by the research team accurately classifies head MRI examinations, enabling automated dataset labelling for downstream computer vision applications.

This agenda item is presented in the following session: S7-4 Radiation Safety and Management

Plenary session

08.10.2022 13:30 - 15:00