Stroke is the second leading cause of death globally. Time to treatment, and thus early diagnosis,
is critical for patient outcomes. AI tools have the potential to help in rapid detection and triaging of infarct and haemorrhage cases
on non-contrast head CT (NCCTB). Until recently, direct comparison between the AI solutions on the market was missing. With
ECR 2023 this has changed, and the picture could not be clearer.
The Need for Direct Comparison
Many AI solutions are available for the detection of infarct and haemorrhage on NCCTB, but head-to-head performance comparisons across AI vendors are rare. At the European Congress of Radiology (ECR) 2023 in Vienna, the European Society of Radiology (ESR) invited four AI companies with AI solutions for intracranial haemorrhage and infarct to participate in a world-first, head-to-head comparison of their algorithms. Analysis was performed two weeks in advance. Results were presented and discussed in a moderated live session in the AI theatre at ECR 2023.
4 vendors participated: Qure.ai, BrainScan.ai, Avicenna.ai, and Annalise.ai (others declined to participate).
4 cases of haemorrhage and infarct on deidentified non-contrast head CT studies.
0 context provided, only CT studies available to participating vendors.
6 hours to process the cases and send back the results.
CASE 1: Dysphasia & right-sided weakness
Acute Infarct
Detected by 3 out of 4 participating solutions
Annalise Enterprise CTB correctly detected a large territory of ischaemia in the left parieto-occipital region. The model also picked up several findings unrelated to stroke, including a small hyperdense lesion in the sella turcica with a differential diagnosis of mass vs. aneurysm. This was possible because of the model’s comprehensive findings coverage.
Annalise.ai
Vendor 2
Vendor 3
Vendor 4
CASE 2: Sever headache
Extra-axial haematoma
Only detected by Annalise Enterprise CTB
Annalise Enterprise CTB correctly detected an extra-axial haematoma, further subcategorised as an acute subdural haematoma. The haemorrhage layering along the tentorium is visible on multi-planar reformats, provided by the model, and highlighted in the Annalise Viewer (axial, sagittal and coronal).
Annalise.ai
Vendor 2
Vendor 3
Vendor 4
The Annalise Viewer provides an intuitive and configurable user interface, automatically correcting for rotation and tilt, reformatting the image, and listing all suspected findings in the Annalise Viewer. Selection of a finding will provide a purple overlay in three planes, subcategorisation, differential diagnoses, and a confidence bar, if applicable. For simplicity, only the infarct and haemorrhage findings are shown here.
CASE 3:
Headache, dysphasia & visual deficit
Detected by all participating solutions
Annalise Enterprise CTB correctly detected an acute intraparenchymal haemorrhage with vasogenic oedema. The model localised the haemorrhage in the left parietal-temporal region. Additionally, the model identified subtle effacement of the lateral ventricle, an arachnoid cyst and a small osteoma in the right frontal bone.
Annalise.ai
Vendor 2
Vendor 3
Vendor 4
CASE 4: Left-sided weakness
Acute Infarct
Only detected by Annalise Enterprise CTB
Annalise Enterprise CTB correctly identified a subtle focus of ischaemia in the right sub-insular region and a hyperdense middle cerebral artery (distal M1 and proximal M2 segments). The model also picked up a vascular malformation and an adjacent focal intra-axial calcification, which was incorrectly categorised by other models as an intracranial haemorrhage.
Annalise.ai
Vendor 2
Vendor 3
Vendor 4
“The Annalise.ai algorithm was the only one that detected pathology on all four cases.”
The Imaging Wire
Results summary
Case
Detected by
Annalise.ai
Patient Clinical Presentation
Ground Truth
1
3 out of 4 vendors
Dysphasia & right-sided weakness
Acute Infarct
2
1 out of 4 vendors
Severe headache
Extra-axial haematoma
3
4 out of 4 vendors
Headache, dysphasia & visual deficit
Acute intraparenchymal haemorrhage
4
1 out of 4 vendors
Left-sided weakness
Acute Infarct
Differentiators for clinical excellence
Annalise.ai takes a no-shortcut approach to developing clinically comprehensive solutions, designed to empower clinicans to make accurate and faster decisions.
The outstanding performance of Annalise Enterprise CTB, as demonstrated in this world-first, head-to-head comparison is fortified by:
One of the largest radiology databases
Annalise Enterprise CTB was trained on 229,000+ unique non-contrast CT Brain studies, including different care settings, multiple manufactures and devices.
No-shortcuts in labelling
At least three fully qualified radiologists independently hand-labelled each dataset, unlike many AI solutions that use natural language processing (NLP) extracted labels. Radiologists had access to future and past context, including MRI imaging and clinical reports, creating a total of over 269 million labels.
Comprehensive approach
Annalise Enterprise CTB identifies up to 130 findings on non-contrast CT Brain studies. Hand-labelling each NCCTB study for a comprehensive number of findings improves overall accuracy by providing extra findings for training that could otherwise be confused with the single finding.
When every second counts, count on Annalise.ai
- WHO global health estimates (2000-2019)
- Feigin VL et al. World Stroke Organization (WSO): Global Stroke Fact Sheet 2022. Int J Stroke. 2022 Jan;17(1):18-29. doi: 10.1177/17474930211065917.
- Lee EJ et al. Impact of onset-to-door time on outcomes and factors associated with late hospital arrival in patients with acute ischemic stroke. PLoS One. 2021 Mar 25;16(3):e0247829. doi: 10.1371/journal.pone.0247829.
- The Imaging Wire Newsletter, 16/03/23
Data presented in this case study is based on the initially submitted results, as presented by the moderator of the ECR session. Subsequent result modifications are not included. This information is intended for health care professionals only. Annalise Enterprise is not intended to provide direct diagnosis. For detailed device information, including indications for use, contraindications, and warnings, please consult the user guide prior to use. Not all features are available in all regions. Check regulatory status with an Annalise.ai employee or contact info@annalise.ai. Annalise Enterprise CTB is not for sale in the U.S. Annalise.ai, ANNALISE ENTERPRISE and the Annalise logo are trademarks of Annalise-AI Pty Ltd, registered in Australia and other countries and regions.
Annalise-AI Pty Ltd., Level P, 24 Campbell Street, Sydney NSW 2000, Australia, Annalise.ai
OPT-PRM-084 V1