Skip to main content

Press release Published on 20.1.2023, 09:38

Artificial intelligence algorithm developed at HUS identifies cerebral haemorrhage from CT scan images

Keywords:
  • HUS
  • neurosurgery
  • artificial intelligence

HUS has been involved in developing a tool based on artificial intelligence which can identify subarachnoid haemorrhage, a potentially fatal disturbance in cerebral blood circulation. The accuracy of the algorithm is excellent, and it is potentially capable of improving the timely diagnostics of patients.

SAV-vuoto TT-kuvassa.png

Photo A shows a head in a CT image with an extensive SAH. Photo B shows areas marked in red where the algorithm has detected blood.

Neurosurgery researchers at the HUS Helsinki University Hospital have been developing an artificial intelligence-based algorithm which can effectively identify a subarachnoid haemorrhage (SAH) from a CT scan. 

A study published in the renowned journal Neurology examines how well the AI-based algorithm works in cerebral haemorrhage diagnostics. Artificial intelligence was taught at HUS based on material from CT scans performed on the heads of patients undergoing treatment.  The functionality of the artificial intelligence tool was also examined against more extensive international data. 

Patients arriving at an emergency room complaining of a severe headache are given a computer tomography (CT) scan to eliminate the possibility of a cerebral haemorrhage. Identifying the cause of a cerebral haemorrhage is important for the treatment of the patient. Up to 75 percent of patients with subarachnoid haemorrhage die within a year of a new haemorrhage if the original haemorrhage is not identified in a timely fashion. 

“CT scans of the head are among the most common imaging studies in hospital emergency care and SAH is the most frequent cause of sudden death connected with disturbances in blood circulation in the brain among people of working age. The AI algorithm was capable of precise detection of subarachnoid haemorrhages. Artificial intelligence could assist radiologists in the interpretation of images by sorting out the CT images which require urgent attention, says Miikka Korja, Chief Neurosurgeon at Neurocenter.  

Diagnoses are still made by a radiologist, and decisions on treatment are made by the attending physician.

The artificial intelligence tool correctly identified 136 of 137 cases of subarachnoid haemorrhage out of a total of 1,300 CT scans.  The material comprised a 49,000 image slices from which artificial intelligence identified SAH in 1845 slices out of 2,110 image slices. 

Research community given access to algorithm

A website has been set up as part of the research article, where anyone can test how the algorithm works by uploading a CT scan of a head onto the site. 

 “Opening the artificial intelligence algorithm transparently for use by the research community is a significant innovation in the field of medical imaging and we consider this type of action to be of central importance for developing models for clinical work based on artificial intelligence”, says Heikki Peura, a HUS doctor and one of the two main researchers of the project. 

The artificial intelligence algorithm which identifies subarachnoid haemorrhages, has been openly distributed for further development. The research team has continued to work in the CleverHealth Network corporate cooperation ecosystem coordinated by HUS in the further development of cerebral haemorrhage algorithms, and the first algorithm package designed for clinical use identifies other spontaneous cerebral haemorrhages, in addition to subarachnoid haemorrhages. The newly published algorithm is part of a new algorithm package. 

HUS hopes to get the package clinically tested in 2023, and also hopes that in the future it can get official approval for the use of the algorithm in patient care as well.  




Original article in the Neurology publication
Antonios Thanellas, Heikki Peura, Mikko Lavinto et al.
Development and External Validation of a Deep Learning Algorithm to Identify and Localize Subarachnoid Hemorrhage on CT Scans.
Neurology 2022;00:1-10. doi:10.1212/WNL.0000000000201710

Editorial in the Neurology publication:

Deep Learning Algorithms for Brain Imaging: From "Black Box" to Clinical Toolbox.