Consistent, unshakeable ringing in the ears is a condition that affects 10-20% of adults around the world. Some 20% of those that have it are afflicted by a severe form of tinnitus which presents with additional symptoms such as depression, cognitive dysfunction, and stress. Up until recently, the diagnosis of the condition has relied on patients providing subjective feedback. Now, researchers in Melbourne, Australia have developed an objective measure for tinnitus, using machine learning algorithms and a process called functional near-infrared spectroscopy (fNIRS.)
“Despite its wide prevalence, there is currently no clinically used objective test that can measure changes in brain activity related to tinnitus,” the researchers state in their paper published in a Public Library of Science Journal this month. “Such a measure would greatly benefit clinicians and patients by providing a tool to objectively assess new treatments and patients’ treatment progress.”
The tinnitus study was undertaken by the Royal Victorian Eye and Ear Hospital as well as the Bionics Institute at the University of Melbourne and researchers from Deakin University. Twenty-three of the twenty-five tinnitus participants had bilateral chronic subjective tinnitus. The fNIRS results from those patients were compared to twenty-one healthy adults — with no history of tinnitus, neurological or hearing disorders – which made up the control group. “In this study, we aimed to apply statistical and machine learning algorithms to fNIRS signals to: 1) assess the sensitivity of fNIRS to differentiate individuals with tinnitus from controls, and 2) identify fNIRS features associated with subjective ratings of tinnitus severity and whether these could differentiate between perceived loudness of tinnitus and annoyance,” the researchers write in their paper. They wanted to understand as much about the results that the AI algorithm reported as possible, and designed the study to maximize transparency. “To avoid machine learning models becoming a ‘black box’ with little information about features and parameters, we have first performed statistical analysis to gain a better understanding of signal features, cortical regions and conditions that show group differences and changes with tinnitus severity levels.”
Testing took place in a sound-treated booth, where participants were asked to sit on a comfortable chair while auditory and visual stimuli were presented to them. While it had been established by previous research that measures of brain connectivity or evoked responses are associated with tinnitus, this study is the first of its kind to analyze the outcomes with machine learning using fNIR. According to researchers, there were three classifiers that showed the highest accuracies when comparing tinnitus participants with controls: Rule Induction, Neural Networks, and Naïve Bayes. While Neural Network and Rule Induction both ‘resulted in accuracies above 70% to classify tinnitus from controls and tinnitus with different severity levels,’ the paper notes that an advantage of using Rule Induction was that it didn’t need as much time to run the algorithm compared to the Neural Network approach. Another advantage of the Rule Induction classifier is that it can “produce understandable rules which may help interpret findings,” the paper adds.
With respect to neuroimaging of the tinnitus condition, the fNIRS method was used to record resting state and evoked responses because it can measure changes in blood oxygen levels in the brain, it has advanced temporal resolution over other approaches, and does not produce scanner noise — making it more suited to hearing-related research. The study authors say the results from this research can now be used to investigate the identification of subtypes of tinnitus, as well as to objectively assess the effectiveness of tinnitus treatments and develop a better understanding of brain networks involved in this condition.
Additionally, research on a larger group of participants is recommended by the authors, as well as repeating the tests after treatment has been administered to measure its effectiveness. “Our findings so far show the feasibility of a machine learning model trained to classify an individual’s fNIRS data to a tinnitus severity level,” the paper concludes. “Repeating recordings after a certain treatment and running these through the same model would be able to highlight changes in severity as a result of intervention.”
Artificial Intelligence is being implemented in industries all over the world and is a central theme of the research undertaken at UCIPT. Our work in the HOPE study is using data to assess and shift behavioral outcomes among HIV and other populations.