AI Improves Diagnosis of Intracranial Hemorrhage

An intracranial hemorrhage is a type of bleeding that occurs inside the skull. When this occurs in the skull, blood pools in the skull, puts pressure on the brain, and deprives it of oxygen. The brain has an emergency energy supply for about three minutes. After this time period, a lack of oxygen supply can lead to irreversible brain damage and death. These events can result many different causes such as head trauma, rupture of blood vessels, or blockage of arteries. Altogether, intracranial hemorrhage is a life-threatening, time critical medical emergency and requires fast, accurate diagnosis.

Diagnosis of Intracranial Hemorrhage (ICH)
There are main four types of ICH: epidural hematoma, subdural hematoma, subarachnoid hemorrhage, and intracerebral hemorrhage. The differences between these types depend on the location of the blood pooling. For example, in epidural hematoma, blood accumulates between the skull and the outermost covering of the brain (dura) whereas in intracerebral hemorrhage, blood pools inside the brain itself. Each of these subtypes of ICH requires immediate diagnosis and distinct treatment plans in order to save a person’s life and preserve as much brain tissue as possible.

The first step in ICH diagnosis is a non-contrast head computed tomography (CT) scan, in which many different X-ray images are combined to create a 3D view of the head. A normal head CT scan looks like the following, where the skull is white, brain tissue is gray, and ventricles are black:

Blood appears bright or hyperdense on a non-contrast CT scans. Different subtypes of ICH can be visualized depending on the location of abnormally bright regions and the changes in the normal brain tissue architecture. Three characteristic images of subdural, epidural, and subarachnoid hemorrhage can be seen below:

AI Diagnosis of Intracranial Hemorrhage
In the past year, there have been several significant studies and companies that showed AI can improve diagnosis of intracranial hemorrhage and can be well integrated into clinical workflows in the emergency room. Here are the details of a few of those studies.


Links to full articles are here:
https://www-sciencedirect-com.ezp-prod1.hul.harvard.edu/science/article/pii/S0140673618316453
https://www-nature-com.ezp-prod1.hul.harvard.edu/articles/s41591-018-0147-y
https://www.nature.com/articles/s41746-017-0015-z

Research Details 

Chilamkurthy et al. trained a deep neural network from a data set of 313,318 head CT scans from 20 centers in India They were able to achieve accuracies (measured by area under receiver operating characteristic curves) between 0.85 to 0.95. This work is the production of Qure.ai (website: http://qure.ai/), who has recently partnered with Teleradiology Solutions and Telerad Tech, was named one of India’s 30 most promising technology startups, and recently won the NASSCOM AI Game Changer Award.

Titiano et al. constructed a convolutional neural network from 37,326 head CTs and developed a triage clinical workflow to flag time-critical, life-threatening neurological events. Their group constructed a randomized controlled trial in a simulated triage environment and showed that their algorithm can interpret images 150 times faster than humans and improve prioritization of acute events.

Arbabshirani et al. created a deep convolutional neural network from 46,583 head CTs from Geisinger to identify different types of intracranial hemorrhage. They were able to demonstrate an accuracy AUC of 0.846, identified additional subtle cases of ICH that was initially missed by the radiologist (4 out of 34 cases), reduced median time of diagnosis (512 to 19 minutes), and improved prioritization of “stat” cases that required immediate management. Their workflow is described below:

Images are internally read by AI and prioritized for further analysis to the radiologist by AI severity.

Application of AI
These studies demonstrate that AI has the potential to make dramatic changes in clinical care by improving clinical workflow through saving time and prioritizing emergency cases to radiologists, showing high accuracy for detection of intracranial hemorrhage, and correcting missed diagnoses

AI Companies to Watch
Two companies that received FDA approval for AI diagnosis of ICH include MaxQ-AI and Aidoc.

MaxQ-AI received FDA clearance for ICH diagnosis in January 2018, and have recently partnered with EnvoyAI, Samsung NeuroLogica, and GE Healthcare. Their website is here: https://maxq.ai/.

Aidoc received FDA approval for their CT in August 2018. They have recently partnered with SaferMD, Antwerp Univeristy Hospital, and Rochester Medicine to integrate their solution in hospitals. Website: https://www.aidoc.com/

It looks like a lot of companies will soon enter this space and it is exciting to see how they will continue to make changes in medicine!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s