A new study published in The Journal of the Acoustical Society of America and led by recent PhD graduate Marzieh Golabbakhsh and Prof. Robert Funnell involved generating simulations of the adult middle ear using data acquired from a novel technique referred to as optical coherence tomography (OCT). This tool, which uses light to record both anatomy and vibrations of the bones of the middle ear, provides a non-invasive means for functional imaging of the middle ear. In this particular study, Marzieh simulated models of healthy and pathological middle ear mechanics and then designed a decision-tree classifier that could distinguish between healthy and pathological models.
We reached out to Marzieh to explain her research and its implications for patients suffering from hearing loss.
Q: First of all, congrats on recently defending your PhD! Could you start by telling us a little bit about your research background and why you decided to join the Funnell lab to carry out your PhD? What interested you about this research topic?
My background is actually in electrical engineering. Before starting my PhD, I worked on speech processing and sound analysis at the Medical Image and Signal Processing Research Center in Isfahan. While doing this research, I got to learn more about human perception of sound because we needed to understand how different patients with speech impediments hear and process speech. It was for this reason that I then applied to carry out a PhD with Dr. Funnell at McGill to work on modeling the biomechanics of the human middle ear.
Q: What’s really new in this research article? How does it add to what was already known?
Unlike the inner ear, which is where neurons within the cochlea are found, the middle ear operates purely mechanically based on the positioning and interaction of three main bones (stapes, incus, malleus) and two joints. Researchers have done a lot of modeling of the middle ear in the past using microCT imaging, which allows you to get static images of the anatomy of the ear and build a 3D model. A recent technology, referred to as OCT, or optical coherence tomography, allows for a non-invasive means of imaging deep within the ear and measuring the vibrations of the structures of the middle ear by inserting a hand-held device into the canal and delivering light. OCT is a bit like ultrasound – you can take static images of the anatomy but also record movement of the bones as they vibrate, essentially allowing you to functionally image the middle ear. In this study, we made use of OCT data attained from cadavers, in addition to microCT data, to generate models of the healthy and pathological middle ear.
Q: Can you tell us more about these models?
We built models based on OCT and microCT data and modified various parameters of the models to mimic a healthy middle ear and three different types of pathology. For example, we created a model to represent when each of the two joints is disarticulated. The third pathology that we modeled involved a stapes bone that does not move. To do this, we made use of high-performance computers to generate simulations with large numbers of models in parallel. This was important because individuals differ in their anatomy and even the various structures have different material properties. By running a large number of models, we could find specific features for classification. We then designed a decision tree using these features that could distinguish between a healthy middle ear and the three pathologies.
Anatomical illustration of the organization of the middle ear
Q: Why are these findings important?
These findings present possible diagnostic applications of OCT measurements, whereby patients with conductive hearing loss can get diagnosed earlier and receive appropriate and timely treatment. The results suggest that the vibration measurements beyond the tympanic membrane (TM) that can be provided by OCT may provide a diagnostic method that is less invasive, less costly, and faster than surgery.
Q: What challenges did you face in conducting this research?
The pandemic made things difficult. I had to log into computers remotely to perform my image segmentation and this was slow due to the remote connection. Additionally, the bones of the middle ear are incredibly small and current automatic segmentation tools do not work well. I therefore had to segment the images manually and this was time consuming!
Q: What questions might other scientists raise about this particular study?
One question that someone could ask is why we didn’t use machine learning in our tools. You can’t necessarily rely on machine learning methods for explanations of phenomena. In my case, I wanted to know what is really happening at the level of functioning of the middle ear – what part of the middle ear is responsible for which type of pathology. I therefore chose the questions and answers for the decision tree based on my observation of the results. In the future, more sophisticated methods can be used.
Q: What are the next steps/future directions for this research?
The next step would be to implement machine learning mechanisms for image segmentation and for identifying specific features that can be used to classify normal versus pathological models and even to generate additional models of pathology. In future studies of pathological conditions, one could include medical history, data from clinical tests, along with the vibration measurements. Synthetic clinical data could be generated to supplement the synthetic vibration data as well.
We employed a decision tree and a limited number of features, namely the vibration magnitudes at the umbo and the incus, at a few different frequencies, to classify normal and pathological ears. However, future studies could explore other classification methods, such as random forests, or other supervised learning methods such as support vector machines (SVM) and k-nearest neighbours (kNN).
Q: How did the McGill experience (or people in the McGill community) help you to pursue/develop your passion for research? Is there anyone here who has supported or mentored you?
Working remotely during the pandemic was isolating, but Prof. Funnell was very helpful. He is also the type of supervisor where, if I had a question, he wouldn’t give me the answer right away but guided me in the right direction, which helped me to learn to think critically. He was also very patient!
Q: Why did you decide to come to McGill?
I like Montreal as a city, I think it’s a great city for students. Also, the Biomedical Engineering Department at McGill is within the Faculty of Medicine and that was something that I gravitated toward because there is a direct clinical relevance to our research.
Q: Is there anything else that you would like to add?
Yes - I am currently looking for a post-doc position in Montreal!