A Forum For All Things CPAP
https://www.cpaptalk.com/
Probably less then what I've seen in the last responses here, but thanks for your valuable input - especially on palindromes, which is absolutely fascinating.chunkyfrog wrote:...Some BS actually gets by.
sleepy1235 wrote:The general press and the technical press is full of information about Artificial Intelligence (AI). I am sure everyone has heard of self-driving cars, robots used in warehouses, robots used to do delivery in hospital etc.
The big thing they talk about now days is Deep Learning and software to do it.
It occurs to me that AI would be great for APAP machines. The software could learn from your sleep and optimize the APAP, or should I call it an AIPAP, to reduce events. In fact it might continuously learn to eliminate events. Deep Learning develops models and decision making models can be very complex and made more complex as AI learns. Also, an AI system could adjust as a person or their environment changes over time.
I read that doctors have to optimize control factors for APAP. That they try different settings and then see how it works over a month or so. There are just a few factors they have to adjust. Some times the factors drive a pressure increase that wakes the sleeper. AI could develop complex models which would be optimal and could have many factors and have very complex curves of increasing pressure preventing an event but not waking the patient.
I think the goal with AI might possibly be zero events for every evening. Since AI learns from failures as well as success it might be as the level gets very low it will stall out in improvement, unless perhaps there can be identified a "near" event to further optimization.
There are papers for using AI for Apnea in diagnosis, but I am not seeing the use of AI for running the APAP.
To have a new technology the first step is to imagine it and why it is necessary.
Yes, control Windshield Wiper control box, $6,000 to replace the old way of intermittent WS Wiper control old way cost $60 seldom broke, what a bargain, as are the Ign controller, they replaced the points or magnetic pulse sensor that replaced it or the ECM computer. We are throwing $100 bills at Ten cent fixes, all we get for it is higher prices and less reliable systems, that's progress, sounds like the Government.jim22 wrote: ↑Sun Mar 05, 2017 7:40 pmI write software for airplanes for a living. We are not allowed to use "self modifying code", which is pretty close to what AI would be. I suppose based on the language used, maybe it would only be data that changed, but the bottom line is that it would be somewhat unpredictable how the machine would behave. Now the machine has physical limits, like it cannot physically create enough pressure to cause real harm. It could, however reduce the pressure below a breathable level, but in that case the masks provide a safety backup vent to prevent harm. To be useful, I think an AI based machine would need access to more information than a basic cpap machine has. If it could know sleeping position, sleep stage including being awake, and who knows what else, it might be able to learn what pressure settings work for different conditions. The advantage is that the machine could set itself up over time in ways that are too complicated and varied for one time sleep studies. I also think that it would not require a separate or expensive update to the processors used in the equipment we have now. There are quite powerful processors that are inexpensive and self contained. We use them to run our cars every day.
Jim
Jim