• Light
  • Dark
  • Auto
Select Page

Traumatic Brain Injury (TBI) is a complex neurological condition that requires fast decision-making in the emergency room. Some lesions need immediate surgical treatment -for instance- to decrease the patient’s intracranial pressure. Taking such a decision is complex for radiologists since it implies the assessment of several variables. Lesion recognition and quantification are mainly assisted by the acquisition of CT images and by their further expert interpretation. Among some of the important considerations, the type and location of the brain damage should be identified.

Conventional vs. complex markers
Recently at icometrix, we inspected which artificial intelligence (AI) characteristics allow the differentiation of different TBI lesions [1]. Through the analysis of a big cohort of trauma patients presenting with contusions, subdural and epidural haematomas on their CT images, a large number of computational inspired markers were computed and evaluated. The analysis included some conventional, ‘visible’ markers, commonly used by radiologists (lesions density, shape, and morphology), as well as more complex computational markers bypassing the human eye (e.g. image patterns, such as texture). By means of machine learning (ML), we developed a computational tool that allows the automatic recognition of different lesion types and helps to detect the most informative markers.

The potential of AI in TBI
The results of this research brought several visible biomarkers to light that aid in the discrimination of lesions, such as mean density, surface to volume ratio and sphericity. Our AI algorithms additionally unveiled computational-inspired markers (several types of texture features) that were as informative as the visible ones for recognising lesions, suggesting the presence of hidden data patterns inherent to the lesion types. Interestingly, the AI system was equally accurate in recognising the lesions with or without the inclusion of the visible markers, suggesting a more crucial dependence on the texture characteristics. In other words, the algorithm learned to conduct the classification task by using different criteria than radiologists.

Our findings confirm the potential of AI algorithms to assist the radiologists in the characterisation of trauma lesions. Future analysis will explore algorithms for lesion detection, delineation, and classification using complex computational markers, in addition to conventional markers.

icometrix (Leuven, Belgium; Chicago, IL, USA) is the world leader in software solutions to obtain clinically meaningful data from brain MRI and CT scans. icometrix will be part of the ECR 2019 Artificial Intelligence exhibition (AIX) in hall X1, booth AI-19.

Text written by Ezequiel de la Rosa, MSc
Ezequiel de la Rosa is a Ph.D. Researcher at icometrix working in the TRABIT consortium (Marie-Curie grant agreement #765148), with expertise in computer science, data analysis, machine learning, and applied medical image processing.

[1] de la Rosa E., Sima Diana M., Vande Vyvere Thijs, Kirschke Jan S., Menze Bjoern, (2019), A radiomics approach to traumatic brain injury prediction in CT scans (accepted for ISBI 2019).

Latest posts

Welcome to the blog on Artificial Intelligence of the European Society of Radiology

This blog aims at bringing educational and critical perspectives on AI to readers. It should help imaging professionals to learn and keep up to date with the technologies being developed in this rapidly evolving field.