Week of 02/05/2018 - Progress Report 10

Tamara Jovanovic

Topics covered: Acquiring maximum stability of the optical system, final reprinting of the cuvette stand, getting glucose data for the first time

Materials used: Solidworks Software, Optical Setup of the Non-Invasive system, glucose, solutions, lab equipment

This week’s and last week’s topics involved acquiring data through our system by using actual glucose solutions. Because of the high sensitivity of the entire system, we moved it further to the middle of the optical table to ensure maximum security and stability. Another thing done to ensure stability is the reprinting of the cuvette stand. After some initial tests, a final version of the cuvette stand was developed. This stand was securely bolted down to the rotational stage and I made certain that it doesn’t move.

After these minor adjustments were made to the entire system, we went back to trying to acquire power at the output collimator. Since the power from the laser rounds up to around 4mW, we wanted to get minimum possible power retention through the second collimator. After some hours spent with they system, we were able to achieve 1.6745 mW of power.

This was a lot of power that we were able to achieve and since it seemed that we were not going to be able to get any more at this point in time, we decided to go ahead and finally start testing glucose solutions.

Another change that was made was that we needed to change from testing glucose solutions in a plastic cuvette to testing them in a glass one. The reason we decided to do this is because at 1550nm, which is what we have from our laser, there is a lot of power attenuation through plastic. This is shown later on, in one of the sections of Ezequiel’s part of the report.

Since the optical system setup was finalized, the next step was making glucose solutions. The same steps were followed to making them as outlined in Progress Report 2. I made six 100-mL solutions with different concentrations of glucose in them. They range from 50mg to 7000mg, so that we can see a trend in the spectral analysis. The concentrations used in each of the 100-mL distilled water solutions are outlined below. The actual, measured amount of glucose in each solution is shown in the table below.

The glucose weights were mixed in separate beakers and taken to the photonics lab for testing.

To ensure that the tests are consistent, I decided to test every solution 3 different times, by emptying the cuvette from the old solution, rinsing it with distilled water and putting the same solution again two more times. Each time would be recorded on the spectrometer and the plots would be generated. Since the spectrometer records every time a measurement is taken, these excel spreadsheets of data will be saved and put into our already developed matlab software for analysis. Three separate traces were used on the spectrometer, to show the consistency (or for that matter, inconsistency) of the testing.

The first thing tested as a baseline was just the spectra in free space. Though there was nothing for the laser to go through, it was a good baseline test to ensure that the results we obtain are reliable. In Figure 1 below, you can see the spectra of 3 separate trials in free space. Though it doesn’t seem like it, there are three lines in the picture for each separate trial. They are just one over another and this confirms the accuracy.

Figure 1 - Spectrometer graph through free space.

Though hard to believe, there is actually three traces on this graph in Figure 1. They are just identical and one over another. As you can see, the trace starts somewhere in the high 30’s range of dBm. Ideally we will work on getting the initial power in free space to a larger number, hopefully somewhere in the 20’s range, but that will probably be another week’s report.

Given this baseline test, the idea is that with every glucose solution and the increase in concentration of glucose, the plot will start at a higher number of dBm’s.

The next thing tested was clear distilled water, also for reference. We got the spectra for free space but since the goal of this project is being able to test glucose solutions, not free space, this test was conducted. Figure 2 shows three trials of spectral analysis in free space.

Figure 2 - Absorption spectra through distilled water.

As you can see from the spectra in Figure 2, the spectra through water already has a decrease in dBm’s, a little below -61, which is a good start. We should expect to lose power through these solutions, and since we already do, that means we are on a good track. If you look close enough, this plot from Figure 2 also shows a plot of three separate trials.

Figures 3-8 show absorption spectra on the spectrometer of each of the glucose solutions. The table below links each glucose concentration to a specific Figure which represents absorption spectra of each separate solution of glucose and distilled water.

Figure 3 - Absorption spectra through a 50-mg glucose solution.

In this figure you can clearly see three separate lines from three trials. Though not exactly the same, this proves that for a 50mg glucose solution, the results are quite accurate. Same goes for the rest of the trials for all other glucose concentrations.

Figure 4 - Absorption spectra through a 80-mg glucose solution.

In the 80-mg solution, the results are a bit more accurate. You can still tell that there are three trials done, they are just more precise and close to each other.

Figure 5 - Absorption spectra through a 160-mg glucose solution.

The absorption spectra is shown in Figure 4 shown above. The trials of this concentration are still consistent, and there is a dropping trend of starting dBm. So far, so good.

Figure 6 - Absorption spectra through a 240-mg glucose solution.

Figure 7 - Absorption spectra through a 2000-mg glucose solution.

Figure 8 - Absorption spectra through a 7000-mg glucose solution.

All the absorption spectra can be clearly seen from each graph and separate trial. The results are very consistent, which is really good progress. All the data from each glucose concentration and free space was fetched and will be analyzed in Matlab.

Plans for next progress:

1. Calibrate the system more.

2. Keep taking glucose data with more calibration.

3. Keep taking glucose data with different concentrations.

4. Possibly take data with a glass cuvette, if obtained over the course of the week.

Ezequiel Partida

Topics covered: Glucose Solution Spectra Analysis

Materials used: Optical Setup of the Non-Invasive system, Matlab, MS Excel

This week, we were all able to align the system properly so we could begin capturing glucose solution data. Additionally, Jonathan was able to acquire a quartz cuvette from the Biology department to see if our normal plastic cuvettes made any difference. Once our system was aligned and we had a hold of a quartz cuvette, the following data was gathered:

As can be seen by the pictures above, there is a significant difference between our plastic cuvettes and the quartz cuvette we borrowed. First of all, there is a massive power loss with even just the quartz cuvette when it is empty (1.6 mW to 130 uW). However, when it comes to the plastic cuvettes we’ve always used, the power loss is extremely significant, giving us only in the nW range. When water was introduced, the power for both the quartz and plastic cuvettes dropped in the nW, however the quartz cuvette still had significant power output, with more than 190 nW. The power from the plastic cuvettes was almost negligible, with less than 30 nW at times. However, even though power was minimal, the Yokowaga Spectrometer was still able to detect visible spectra, as shown in the last picture. In order to differentiate the different spectra, I plotted the Spectrometer data for those spectra in Excel, as shown below.

As it can be seen above, the 2 spectra of the plastic cuvette have a very different shape and resolution from the top Free Space spectrum. This is bad, because when taking glucose solution data, the noisy spectra and unreliable shapes will give us unreliable analyses. However, the orange spectrum, Quartz Cuvette Empty, is significantly similar to the Free Space spectrum. Almost no shape difference, and only a slight power loss visible. Moreover, the 2 middle spectra, 1 with Quartz cuvette with water and the other 1 with Quartz cuvette with glucose water, also don’t lose the spectrum shape too much. The spectra are also consistent. Therefore, for this week, when moving on to taking glucose solution data, we decided to use the Quartz cuvette since it generates better preliminary results.

After Tamara took multiple glucose solution data, I wrote a Matlab script to sort the data, plot the spectra, and perform preliminary analyses on them. Tamara used the quartz cuvette to capture all data. She took data for free space spectrum, a spectrum with distilled water, a spectrum of 50 mg/dL solution, 80, 160, 240, 2000, and 7000 mg/dL solutions. For each of those, she took 3 different trials. Below is what I did with all her data (The first picture shows that for each type of spectrum, the 3 trials were saved in their own individual excel files, directly coming from the spectrometer. The excel files were used in the Matlab script, shown in subsequent pictures).

These plots show that the 3 trials for each plot are significantly similar. Therefore, an average of the 3 trials was taken for each type of data in order to perform analysis.

By zooming in, it can be seen that the different glucose solutions are all less power than the reference spectrum with just distilled water. However, as analysis will show, the differences between the solutions are not consistent with the concentrations.

As can be seen, the correlation for each of the glucose solutions with respect to the distilled water reference was always basically 1. This means that our correlation method accurately detects the proper shape of the spectra. However, the RMSE% showed different results for the glucose concentrations. The 50 mg/dL solution, as expected, had the lowest percent error from the distilled water spectrum. However, the next lowest was the 240 mg/dL solution as opposed to the 80 mg/dL solution. Asides from that, however, the rest were as expected. Additionally, when performing these analyses, different sections of the spectra can be analyzed by controlling the length and sections of the data vectors. By getting the first 80 points from each of the data, the following results were gathered.

This time around, the most similar spectra were the 50 mg/dL and distilled water spectra once again. However, the rest of the solutions were more mixed up than before.

Therefore, this week we discovered several things. First of all, it is evident that the plastic cuvettes absorb too much power and generate unstable spectra. The quartz cuvettes are much more reliable. However, we might get away with regular glass cuvettes. It just a matter of testing out different materials and taking more tests. Also, we found that taking glucose solution data is possible. The resulting spectra are correlated well with the free space and water reference spectra. However, we also found that with these preliminary tests, different concentrations seem to generate unstable spectra and subsequent analyses. There could be several possible solutions. First of all, taking 3 trials for data might be too little. Perhaps, characterization of data has to be performed, taking tens, or hundreds of data sets per type of data (i.e. water spectra, different solution spectra, free space spectra, etc.) in order to get a reliable reference for each. Then, we have to characterize what section of the spectra generates better results for differentiation of concentration (e.g. 1513-1523 nm vs 1513-1660 nm ranges). Also, perhaps our data acquisition strategies are not reliable. Maybe we need to be more careful with the cuvettes and solutions, and we need to have more practice. However, for this week overall, it is evident that glucose solutions data is viable. We have high hopes for the weeks to come.

Plans for Next Week:

1. Continue taking glucose data.

2. Get references for each type of spectra by taking tens (or hundreds) of data and taking stable averages.

3. Determine what section of the spectra gives better analyses results.

Conclude next steps: Do we need bigger wavelength range (i.e. supercontinuum generation)? Do we need more power output? Do we need more data? Do we need better practice and techniques? Etc…

Jonathan De Rouen

Topics covered: Power retention of optical system and analysis

Materials: Plastic Cuvette, Quartz Cuvette, Spectrometer

This week in order to improve our system I worked on finding a new material in order to hold our samples. Due to the absorption of the plastic cuvettes we were unable to obtain sample data that would be able to be analyzed and assessed. I reached out the the biology department and talked to Dr. Yong-Jun Li, manager of laboratory facilities, about the absorption of the plastic cuvette. From our discussion, the plastic cuvettes we were using were not appropriate for the analysis of glucose at our wavelength of light. Plastic cuvettes were found to allow light to pass through 380-780 nm of light or the visible spectrum of light. We were working with NIR light and as such this cuvette was unusable since it would not allow light to pass. However, thankfully Dr. Li was able to acquire a Quartz cuvette for our project to borrow as shown below.

The Quartz cuvette allows light to pass from 190 to 2500 nm. This was confirmed to pass through, and provide better data by using the spectrometer.

In the spectrometer image above the yellow line is the laser through free space. The pink graph below that is the laser through the quartz cuvette. Compared to the lighter pink graph second from the bottom which was the laser through a plastic cuvette. There is much more information gathered and able to be used from the quartz cuvette than the plastic cuvette. The green graph is the quartz cuvette filled with distilled water. This can be compared with the blue graph on the very bottom which was the plastic cuvette with distilled water. Finally the gray graph was the quartz cuvette with distilled water and a random amount of glucose. So from this test it seems that we can finally start testing our measurement algorithms for glucose.

Although the quartz cuvette passed the light because of it being a loaned item to us I wanted to get something more concrete and looked also into using glass cuvettes as a cheaper source. The glass cuvette should also work with our wavelength since they also pass from 340-2500 nm wavelengths. However I was advised to find a glass cuvette in the bio department to borrow to confirm prior to making an order for a new cuvette.

Plans for next week:

1. Borrow a glass cuvette and confirm if it can pass

2. Create a decision matrix and try to find a cheap cuvette that will work (due to our budget)

3. Assist Ezequiel in analyzing the data directly, Possibly return to researching machine learning, if the data is still too sensitive for our proposed algorithms