 Hello everyone, I'm going to share the work that we are doing at the Bionics Institute using functional neuro-infrared spectroscopy to accelerate early intervention for infants with hearing loss. First I'd like to acknowledge my team at the Bionics Institute and also Dr Jeffrey at New Genie, our new company, and our prototyping partner Design Plus Industry. So our aim is closing the gaps between identification of hearing loss and the accurate fitting of the optimal hearing instrument. So where are the gaps? In auditory neuropathy, what is my functional hearing like? For all infants, is my hearing aid programmed optimally and would a cochlear implant be a better option? And for cochlear implants we need objective programming and evaluation of early programs. So what do we need to fast track early intervention? And we need a reliable and accurate objective test that addresses those gaps and is suitable for very young infants in the natural sleep state and suitable for use in cochlear implants. So introducing ear genie, which uses F-nears. So I think most of you are familiar with F-nears now, so basically we're looking at the hemodynamic response in the brain. So we're looking at changes in the oxygenated and deoxygenated hemoglobin in the blood and response to neural activity in the brain. Here we have a picture of a baby fast to sleep having her hearing tested with one of our early prototype devices. So how does the test work? There are two tests. First is a detection test. It's basically a simple block design where we play five seconds of a stimulus that we want to test whether the baby can hear. And so we might play bar sounds, bar, bar, bar for five seconds with silence in between the blocks. And for the discrimination test, it's essentially the same thing except that we have two sounds. So the standard sound is played in the background between the test blocks. So all that gray region is a repeated standard sound. It might be bar, bar, bar, bar. And the test block contains the contrast sound that we want to see whether the baby discriminates it from the standard sound. So it might be gar, bar, gar, bar, gar, bar in that test block. So in this case, we're just looking for a response to the test block. We're using optodes in the auditory and prefrontal regions. We expect to see responses to speech in all of those regions. So here are the average responses for normal hearing infants to the detection of bar at 65 dB SPL. So you can see that the red and the blue lines go opposite each other. That's what we expect to see with hemodynamic responses with the oxygenated and deoxygenated. And we see that there's two components. So there's a positive going response that we can see that it looks like a standard hemodynamic response to a sound. But there's also this huge negative dip here. So we investigated this morphology and we found that there are actually at least two responses. We did ICA on all of the responses and what we saw were two main components. The first component, this red one here, looks like a standard hemodynamic response in the auditory system to sounds. But there's this other big blue response here, which we deduce is due to arousal of the brain. So we're not waking the baby, but the brain goes into a different arousal state when we present sounds. So the responses to the discrimination of bar from other sounds, they look fairly similar. Once again, there's positive going bits and there's negative going bits. So here the three different colors are three different contrasts and they're the HBO and the dotted lines are HBR. So we see the same negative arousal component in many of these morphologies. And we see that the negative component is quite different in the three different contrasts. And we also see that the positive component can be have quite late latency and is quite broad compared to the positive response in the detection response. So how do we statistically test whether a baby actually heard or discriminated a sound? Now, this is tricky because we need to be able to automatically detect non-stationary responses. So both the negative arousal response particularly is not there all the time. It habituates very rapidly in some circumstances and not others. And the positive response can have different types of latency and duration. So we need to be able to detect these. And the standard statistical methods for FNAS don't work. You'll never find anything that works because we don't know what the response looks like and it changes with every epoch that you test. So what do we do? This is company secret. But in very brief, we do standard pre-processing steps to clean the signal and convert writing intensities to HBR and HBR. And then we use a model which uses a stochastic process to capture unique statistical properties of the neural responses. And then we establish statistical significance by comparing the derived neural stochastic process against arbitrary baseline signals. So the baseline is derived from silence periods or in the discrimination test in the non-silence background. We also do another test to test for the probability of false positive responses. And this is important because you can make things as sensitive as you like and the more sensitive you make them, the more false positives you're going to get. So basically to test for false positives, we compare the baseline responses with other baseline responses using the same statistical model as we used to do the detection. So what are the results we get in normal hearing babies? So the top table shows you the number of infants out of 32 infants who showed statistically significant responses in numbers of regions of interest. So the majority of babies showed significant responses in three or four of the four regions of interest. And only two babies didn't. And no babies actually didn't show a response in any of the regions. So a similar pattern for discrimination. Most of the infants showed significant discrimination responses in at least three or four regions of interest. And only three individual tests showed no response in any region here. So in the false positive test, none of the actual false positive tests came up with a false positive alarm. So we're estimating the specificity of the test at 100%. So I'm going to show you two examples of babies with hearing loss who we've tested. So this first example is a six month old infant with moderate hearing loss. And we tested her in the aided condition at 65 dbs perl just in a single ear. It's a single ear response. So on the left you see her detection responses. If you look down here in the temporal regions, you can see that the temporal response looks very, very similar to the normal hearing baby detection responses with positive going and then the negative component of it. Over here in the discrimination responses, you can see that they're mostly apart from Ga, just have the positive component going component. And the positive going component is much later than the positive going component here. So these responses were all clearly significant using the analysis algorithm so that we know that this baby's hearing aid is adequately programmed for her. And that she can also discriminate the speech sounds very well. And so that's really good news. So this is another example of an infant. This infant's 15 months old with developmental delay and has profound hearing loss with a cochlear implant. There was some question about whether she could hear with the cochlear implant because she was extremely difficult to evaluate behaviorally. So on the left, you see at least three of the regions of interest a very large positive response. It's actually much later than a normal detection response, but it's actually there. So we know that the baby is at least hearing the sounds at 65 dbs perl. The discrimination responses are very messy and noisy. And but the algorithm was able to actually statistically show statistically that she could discriminate some of these responses at least in one or two regions of interest. So finally, I'd like to show you our progress with making an actual clinical system. So here we have a conceptual design made by our partner's design plus interest. It's very baby-friendly. It has a fixed montage and a Velcro closing. So it just fits snugly around the baby's head. It has a wireless connection to a computer, which has a clinician-friendly user interface. And it outputs an automatic analysis and report. So we'll be heading into clinical trials with this new prototype towards the end of this year. And I hope you can all participate by sending us babies. Thank you. So thanks for listening. I hope you enjoyed this presentation.