New Method to Map the Inner Ear Using Artifcial Intelligence (AI)

Artificial intelligence helps doctors address inner ear problems

MAASTRICHT / ANTWERP, February 18, 2021 - A new method to map the inner ear using artificial intelligence (AI). That is what the Belgian start-up Radiomics has developed in close collaboration with doctors and scientists from Maastricht UMC +, Maastricht University, Antwerp University Hospital (UZA) and the University of Antwerp.

Researcher Akshayaa Vaidyanathan of Radiomics explains: “To examine the inner ear with MRI scans, you have to be able to recognize the inner ear very precisely on such a scan. That is quite a challenge with a small structure. Until now, doctors manually indicated exactly where the inner ear is located on a scan, but this is not only very time-consuming, it can also lead to large differences between them.” To solve this problem, a new method was developed based on artificial intelligence. This application makes it possible to recognize the inner ear fully automatically on the MRI scan. Not only can this be done within seconds with this new application, it is also possible to do this over and over again with a particularly high reproducibility. ENT doctor Prof. Vincent Van Rompaey of the University Hospital Antwerp: “This application can help doctors in the future in the diagnosis and treatment of disorders in the inner ear, but also be helpful in training doctors.”

Labyrinth
Deep in the ear is not only the cochlea that makes us hear, but also the vestibular organ. Together they form the inner ear (or labyrinth), located in the petrous part of the temporal bone, the hardest bone in the human body. We hear and maintain our balance through a complex interplay of pressure waves in the fluid in the inner ear. Vibrating movements of the air (sounds) are amplified to such an extent that a pressure wave can be generated in the cochlea to activate the hair cells and the auditory nerve.

Head movements set the fluid in the balance organ in motion, and then activate another type of hair cells. The equilibrium nerve will therefore be able to control the eyes at lightning speed and keep your image sharp while moving.

If the complex interplay of pressure waves is not balanced, it can cause hearing loss, ringing in the ears or dizziness.

Teaspoon
To better understand certain inner ear diseases, it is important to be able to properly examine the inner ear. But the inner ear is very small and is therefore difficult to examine. ENT doctor and balance expert Dr. Raymond van de Berg of Maastricht UMC + makes a comparison to make this clear: “A teaspoon (5 milliliters) can contain 25 times the content of the inner ear. This new technique with artificial intelligence can help us to map the inner ear much better and faster with an MRI scan.”

Link: https://www.mumc.nl/actueel/nieuws/...elpt-artsen-problemen-binnenoor-aan-te-pakken

 

"ENHANCE!"

(I had an MRI recently and I've been waiting for an excuse to say that.)

Could this be used on MRIs that have already been taken prior to the technology being announced? It would be great if those of us with existing MRIs could have the inner ear evaluated using this.

 

Is this just a way to examine a structure most radiologists ignore because it's hard to examine and they aren't trained to examine it or is this actually getting new images? It wasn't clear from this.

 

I do not have a clear idea that radiologists aren't generally well trained in examining the specific area of the inner ear, but it may seem that it's hard to examine inner ear pathologies using conventional MRI. The authors of this study argue that their AI method allows them to see more in-depth of deviant morphological shapes in the inner ear, which may be related to inner-ear disorders. Here's an excerpt of the clinical implications that are mentioned in the study of Vaidyanathan, et al. (2021):

Clinical implications and future perspectives

The future clinical advantages of automated 3D image segmentation of the inner ear are versatile. Image segmentation can be used for 3D visualization, allowing a better understanding of the spatial relations and morphological changes within the inner ear, assisting radiologists in the diagnostic process and providing tools for surgical planning or learning purposes. Previous studies have proven the usability of auto-segmentation for pre-operative planning of cochlear implant surgery using CT imaging and for the diagnosis of adolescent idiopathic scoliosis using MRI imaging. Our model proved to be efficient on MRI imaging. However, the proposed methodology can be easily leveraged for similar auto-segmentation applications on different imaging modalities.

Nowadays, quantitative analysis of the inner ear is gaining more importance. Techniques like radiomics, volumetric assessment of fluid compartments in the labyrinth and the analysis of the morphoanatomy for the vestibular system are used to aid diagnosis of vestibular diseases. Radiomics refers to the process of the automated extraction and analysis of large amounts of quantitative features from medical images. These features are sometimes not perceptual for the human eye and might contain information that reflects underlying tissue heterogeneity and pathophysiology. Quantitative image features involve descriptors of shape, size, volume, intensity distributions and texture heterogeneity patterns.

A histological feature strongly associated with Meniere’s disease is endolymphatic hydrops (EH), a distention of the endolymphatic compartment in the inner ear. In conventional MRI, the endolymphatic compartment cannot be distinguished from the perilymphatic compartment, and thus, EH is not depicted. The differences found in radiomic features between MD and controls could hypothetically be explained by the different composition of the fluids in the labyrinth, causing a different distribution of signal intensities. Possibly, EH is captured in the quantitative image features due to damage to or morphological changes to the endolymphatic space. Since Meniere’s disease is still a clinical diagnosis challenge, discovering distinctive image features might benefit the diagnostic trajectory of MD. Another study showed that cochlea CT image features can be useful biomarkers for predicting sensorineural hearing loss in patient with head and neck cancers which received chemoradiation therapy. Different machine learning methods were used for feature selection, classification and prediction. The advantage of using machine learning in combination with radiomics is that the analysis of the labyrinth could be done autonomously in the future. However, for both studies, setting a Region Of Interest (ROI) by manual segmentation was necessary. The fully automated segmentation of the inner ear contributes to efficient research on quantitative image analysis of the inner ear.

Link: Deep learning for the fully automated segmentation of the inner ear on MRI

 

Well, according to the study, these MRI scanners already exist. You only need to have a specific kind of MRI scanner that allows manual segmentation.

 

Thanks for the info.

Off hand, this seems similar to what they are doing with mammograms now (using machine learning to make more objective calls on subtle findings that might otherwise be more subjectively evaluated) but with the inner ear, in particular hydrops.

Seems like a good idea for sure, especially if the difference on MRI between perilymph and hydrops is a very subtle granularity difference.

 

But this still won't allow you to see if the hair cells in your cochlea have been damaged, right?

 

MRIs can't see microscopic structures so that's correct.