First, we have to distinguish between models built for research purposes and models that we built for clinical purposes. SLEEPYLAND is for research purposes. All the models we usually build are for research purposes. Why? Because it’s really and super hard to then convince or let the user, let the scorers use your model, the model X, the model Y. And the reason behind that is because a single user, a single scorer usually uses a specific software...
First, we have to distinguish between models built for research purposes and models that we built for clinical purposes. SLEEPYLAND is for research purposes. All the models we usually build are for research purposes. Why? Because it’s really and super hard to then convince or let the user, let the scorers use your model, the model X, the model Y. And the reason behind that is because a single user, a single scorer usually uses a specific software. And the whole procedure starts from maybe looking at the data on your own software. So you should be able, if you want to use a model that we develop, for example, first of all, maybe to ingest this model in this software and this would be hard because you have to somehow agree with company X or the company Y and so I mean I will not go through this. But the other point is that maybe the other solution would be that you should be able to satisfy the scorer in terms of so you should feel all the steps that we go through the scoring so I’m using my software I should export the data from my software I should then give this data to the tool that you want to use and the tool itself should give you a turn back so give you an output answers that you as a user will be able to import in your own software. And so if not I mean if we do not satisfy all these steps if we do not fill all the steps the user will not be able to use it in a clinical routine, in their clinical routine. And then, I mean, not even mentioning all the ethical problems, other problems that you would have to run this kind of solution or even to share your data with, I mean, if the tool was not local, you had to share your data somehow or upload your data somehow in another server. So I think we are far, let’s say, away from introducing this kind of software. But again, these are tools that have been developed for research purposes. I mean, in a clinic practice, what you need is to… At the end, you want a hypnogram. At the end, I’m just talking about sleep staging, not sleep scoring as a whole, you know, arousal detection and everything. We’re talking about sleep staging, and at the end, you want a hypnogram. But first of all, the hypnogram that you would receive in output from a sleep staging model would be different, of course, based on the model you would use. And in general, the scorer just needs at the end to compute, not just need, but they need the sleep stages. You will compute the macro structure, et cetera. So you will go through this procedure that is much more close to what you would need in a clinical setting. When I say research purposes, I mean, when I discern the tools for research purposes, for a clinical purpose, it means, I mean, for example, for research purposes, it would be super useful to have a representation like the hypnodensity that I mentioned before, I mentioned before, because it’s much more richer than the output that the representation that the scorer would, it’s used to look at. It’s the hypnogram. So I think that the models that we have are already good enough to extract the hypnogram that you would like to see. But it’s also true that there is not a single hypnogram, correct hypnogram, out there. I mean, it’s not about one sleep stage at a time, one sleep stage each of the 30 seconds we’re used to windows. But it should be considered much more dynamic. But the kind of dynamics is not in the clinic at the moment. This is not the way we, at the moment, I’m an engineer, to be honest, but talking with people working in a clinical setting, they actually, at the moment, we have the AASM manual and the rules are set in at least 30 seconds, so the dynamic is not there at the moment.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.