electronic nose circuits

Specificity is the major hurdle in chemisorption sensors. Readout is much less of an issue.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

Aha. You should join the neural net mystic seer clan. :) In more down to earth terms, a bank of fairly non-specific sensors can differentiate "situations," as long as there are enough of them. Even multi-variate dimensions with large SD can differentiate sub-sets. Of course the phenom should be multi-Dimen.

AN analogy I use is how a jury declares someone guilty based on vague circumstantial evidence. So Scott Peterson did 8 vague things, each a little out of the ordinary, that added up to his guilt. In the same way, a statistical post-processor on the "enose" can separate diagnostic situations, like cancer vs.non-cancer. It's attracted some attention, because it's non-invasive and works as well as biopsies in some cases. j

On a sunny day (Mon, 10 Mar 2014 06:44:11 -0700 (PDT)) it happened snipped-for-privacy@gmail.com wrote in :

IS IT NOT EASIER TO SNIFF THE COKE YOURSELF?

Interesting premise. I've heard that the best electronic sensor is still

1000X times less sensitive than olfactory, but that was old information. A lot of work has been done recently in the ?? areas with focus on dangerous 'odors'.

The premise is accurate. illness can be smelled. Not going to make a lot of progress there, until, and if, we can develop a vocabulary describing smell. Just like, mankind didn't make much progress in communication until mankind developed a vocabulary for 'feelings and thoughts'.

In answer to your first question, which "sensor to signal" converter technique to use? Why are you not using them all? You mention how multiple sensors pull out information, then seemed to dismiss the advantage of using all three. The techniques can't be that exclusive to use. Are they?

LOL! envisioning a robot junky!

I did walk into a crack house in Detroit and advised them to avoid sugar and take their vitamins. I beat it out of there when I saw a Crack Ho approaching, as I have my limits.

hmmm. The sensors are just sensing various gases like acetone vapor. The Chinese doctors have a vocab of this, but my conceit is to let the bots do it. For example, I can't sense smells like a dog, or even remotely close to it.

Just like, mankind didn't make much progress in communication until

That's an interesting statement. It's a long discussion, and I wonder if it's true, not just being argumentative. Buddha (600 BC) immediately comes to mind. He taught that emotions were evil, and threw a monk out of the order who disagreed with him. I've spent quite a bit of time with mystics in India, and Tibetans. The latter have a unique vocab of consciousness, but the vast majority of humans are sleep-walking, because they are unaware of the context of their awareness.

Yes, they are, in the sense that some are very laborious to set up. Part of the reason I talk so much is I am lazy, and looking for a magic easy solution. Well, actually I am smart, since I know how difficult the sensing part is. You need some sensor very sensitive, and I'm not sure at this point what that is. Is it spectroscopy? interference? TIR evanescent wave? laser speckle reflectometry? ionization effects? dielectric sensing? current leakage? and so on. It gets a bit counter-intuitive, since we are not seeking differentiation necessarily, just sensitivity mainly. The differentiation has to be there, of course, but it is secondary to sensitivity. At east this is the philosophy of having a pattern recognition layer behind a bunch of sensors that differently sense a multi-dimensional entity.

So I'd like it to be easy to do, very sensitive, and modifiable.

Another issue is "How much can signal processing encroach on the domain of the sensor?" I mean the immediate DSP right after the sensor, not the pattern recognition layer at the end? To what extent can lock-in and signal averaging pull signals out of the noise, that are useful in this context? And if you don't a priori know what and where they are, what s the search algorithm for them?

Open questions. jb

Neural nets are toys, by and large. The main reason is that you don't know any given one works in any great detail, so there's no way of knowing how it will behave with inputs outside its training set.

Statistical learning, e.g. data mining, is a more fruitful approach for that sort of job. (I'm doing a couple of transcutaneous blood spectrometers just now, and there are all sorts of issues of that sort.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

People do NOT know the power inside themselves! Examples are everwhere. Major problem is that we're taught from childhood to ignore, or that such capabilities can't exist, crippling too many useful abilities. It boggles the mind what would be possible if a child were raised without those biases and their minds allowed to form all those related synapses up through the age of five.

True, the arrogance of the West assumed nobody thought of it until 'they' did. Didn't Indian mathematicians find the earth round and estimated its size, circa 0 AD?

Matter at hand...did you do a search into the SBIR program of grants and published results? For several years the govt has been funding development of specific search electronic 'noses' The sensor technology may be public domain and of use to you.

Search algorithm, my premise is that when a closed solution does NOT exist. Assume something and then prove it fits, or does not. Kind of like solving those really sticky differential eequations, where you assume a solution, it works, you're done! Doesn't even have to be a unique solution, either.

Yes, it gets philosophical though, at the extremes. If you look at the literature on biomedical diagnostics, you will see many people using the NN's. If you are smart enough to make data mining work, then i'm very glad to hear more how you are using it, as much as you want to say. The issue with neural net training sets vs. real life is dealt with by holding out part of the training set and testing against that during training. To be fair, any statistical system you use will have the problem of "over training," particularly the multi-dimensional guessors. Another problem, say with transcutaneous NI Hb systems, is the "ecological fallacy," which just states that an individual can have some skin condition that throws off your measurement. Without borrowing trouble necessarily, you are probably looking for something definitive in the spectrum. In general, my comments were directed at horrendously multi-dimensional problems like "Does this person have lung cancer?" That you can get any definitive answers from a series of simple measurements is pretty amazing in my book. Did you get that spectrometer going? I am waiting on a OS for my XP PC with serial port... JB

snip

Here's how you can data-mine to answer question of importance of NN's and medical diagnostics. I went to Google patents and did these searches:

"data mining" "medical diagnostics" - 1400 hits "ANN" "medical diagnostics" - 10,500 hits

ANN stands for artificial neural networks.

That being said, data mining is extremely generic, and a different type of lizard from ANN's.

jb

No, it was known far further back than that. The Pythagoreans noticed that the shadow of the Earth on the Moon was always round, and the only body that always casts a circular shadow is a sphere.

A Greek mathematician whose name I forget--maybe Eratosthenes or Aristarchus of Samos--measured the diameter of the Earth by noticing that at Syene in Egypt, if you looked down a deep well at noon on the summer solstice, the Sun reflection was right behind your head, meaning that it the Sun was directly overhead.

He arranged for somebody to make the measurement of the shadow angle of a vertical wall at (iirc) Alexandria, which was a few degrees. From that, he calculated the Earth's diameter at about 18,000 miles, which is pretty good for such a crude method.

The Earth has been known to be round by everybody who cared, for a good

2500 years. (Yes, including medieval Europe. They read Ptolemy.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

In the literature, sure. They're great for grinding out grants and doctorates. Do you know of a FDA-approved medical instrument using neural nets?

Super secret sauce.

But since you know how it works, you can calculate how sensitive it will be to any of those effects. Not so with neural nets. Big difference in the confidence level.

Yes, it's in clinical trials for both glucose and alcohol. Still some motion control issues, as I talked about last week, but it looks like it works OK.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

I mean _circumference_ at 18000 miles, only about 30% off, which is still pretty good nowadays for an astronomer. ;)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

Well, it may work..for a while...until it essentially saturates and possibly becomes impossible to remove contaminants. Naturally, sales does not want to get even close to these problems...

For the medical profession, the eNose should probably also smell money. Nice replacement for having the doctor do a biopsy on my bank account.

Warning, topic drift ahead...

I had a problem a while back. I came back from a hot-chocolate break to find the office filled with the smell of overheated electronics. The problem was that I couldn't tell where which device was trying to burn down my office. I usually shove a vinyl hose into my nose and sniff for the source, but this time I had a better idea. I have a Nikken AQM (air quality meter). My guess is that it's optical and looks for light scattering from smoke particles. However, I'm not sure as I haven't torn it apart yet: I fired it up, and it instantly displayed full scale. The office was full of smoke. So, I turned on the HVAC fan thing and cleared out the room air. After some random sniffing, I found that an APC UPS had decided it was time to burn up. Nice. I wouldn't mind having a miniaturized version of such a meter for sniffing out individual smoking components on a PCB board. Yet another product idea.

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

Maybe some time ago but no more, and certainly not in medical diagnostics.

Jeff, ur singing to choir there. It should be noted, though, that one approach is "creative disruption" of the medical system. Some docs (Eric Topol) think that we should have our own personal diagnostics to replace the medical system altogether. (a bit of an exaggeration.)

Since I am a biochemist as well as a technologist, I have been around various medical innovations in the past, and seen them suppressed. This is especially apparent in the field of diagnostics. Diagnostics are classified by the FDA as "most dangerous medical devices." Type 3 as I remember.

For example, the PSA test has been greatly improved on to definitively state whether you have prostate cancer, rogue cells, or just BPH. An acquaintance spent some 20 years improving the PSA test. He was a Professor at an institution you would recognize, and he had all his ducks in a row, scientifically and technically.

But because of regulatory road-blocks, we do not have the superior test. (BTW, the FDA tried to suppress the original test, and did, but had to "lead from behind" when docs started using it off-label. What a pun.)

My point is, with a situation like this, don't imagine that you will see the medical innovations you may read about. Currently, there is some indication that certain DNA/RNA manipulations can extend life span to 150 years. Expect that these and other technologies will be suppressed.

Suppose you develop a simple diagnostic for diabetes, for example - it could be non-invasive. It is forbidden for you to sell this test in the "advanced" countries like the US. FDA with police powers can break down your doors, arrest you, and imprison you.

Jeff, how do you like Santa Cruz? I used to live there. A unique community. JB

Interesting article, but it's an illustration of my point, not a refutation. To get specificity, they had to work pretty hard. It's the magic goo that's the difficult part, not the readout mechanism.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net