Role of Optical Spectroscopy in Diagnosis of Prostate Cancer

Role of Optical Spectroscopy in Diagnosis of Prostate Cancer

This presentation is slightly different from what we have seen before. It’s applying engineering methods and techniques for prostate cancer diagnostics, basically bringing something from the bench all the way to the clinic. The light interacts with biological tissue in a number of different ways scattering, fluorescence, Ramen scattering, etc. There are differences between these interactions when they hit benign versus malignant tissue. The idea is if you can quantify these differences, then you can diagnose the disease. In these regards, we have developed an optical biopsy needle for prostate cancer diagnostics and our present basic bench results. Also we have run a pilot study, first in human clinical trial. Dr. Crawford conducted the clinical trial at Inner-City hospital, data from that trial as well.

The main need of this device is because almost 90% of the biopsy cores we currently obtain from TRUS biopsies, as well as mapping biopsies are benign. So, how can we increase the diagnostic ease of these biopsy cores? There’s a way to make the current needle smarter so we can see how it works out. I have a disclosure. The technology has a U.S. patent given to the University of Colorado and has been licensed by Precision Biopsy. We have capital investment from a private equity group, Allied Minds in Boston. The CEO is Mr. Amir Tehrani since 2011. I’m one of the founding members and so are Dr. Scott Lucia and Professor John Daily in the CU mechanical engineering department.

First question, just wanted to ask you to see whether you know some of these things regarding fluorescence spectrum. These are naturally found in your body. See if you can answer the question, which one of these will fluoresce? These are proteins and enzymes. When the light hits, which one of these will fluoresce or emit the fluorescent signal? The answer is all of them, including fatty tissue. That is going to be a confounding condition when you try to use this technique. All three, amino acid enzymes, fatty tissue, they all fluoresce.

The next question, if you use scattering, the light hits these tissue cells then can we use the scattering signal for what type of diagnostics? Benign versus malignant, benign versus pre-cancerous lesions, the histopathological grade of the disease. What do you think? Yes, all of the above. The correct answer is all of the above, thank you.

You’re well informed about the disease and about the prostate cancer. There’s going to be about 233,000 new cases this year, about 29,000 deaths, and the cost of treatment is about $3 to $4 billion. The cost of screening is about $2 to $3 billion for prostate cancer. The dilemma is more men die with this disease rather than from it. It is important to diagnose the so-called aggressive lesions. The screening is based on PSA and digital rectal exam. The diagnosis, still based on prostate biopsies, is the gold standard. It’s about 50% sensitive. The main issue is most of these cancers that are detected will never be life threatening.

The question of what treatment becomes a huge issue. We currently use TRUS-guided biopsies to diagnose this disease. The problems are these biopsies are taken randomly. We don’t know the tissue morphology. We just insert the needle and take a sample. Because of that it is subject to sampling errors. The current clinical detection rate is somewhere out in the range of 25% to 35%, and about 50% of these clinically important cancers are missed. The random biopsies fail to diagnose the so-called significant lesions, and most of the time, fail to provide the accurate grade and the stage of the disease.

As I said at the beginning, we can use optical spectroscopy to get an idea about what type of tissue it is you’re going to biopsy before you take a biopsy core. Light interacts with these biological tissues in a number of different ways, and the optical properties of the tissue are dependent on the molecular composition and the cellular morphology. Due to the malignant growth, there are some changes in these things. Of course, quantifying these things in real time is the challenge. You can put a probe in and quantify these things in real time. Then you can provide feedback to the doctor saying, “Okay, you’re close to a malignancy,” take a biopsy, then that will be very valuable.

Fluorescence is basically where you have a photon, then the photon is absorbed by these electrons, and they go into an excited state. Then when they calm down, they emit a signal. Most of the time that’s what you call a fluorescent signal. These are not necessarily in the visual range. You cannot see these with your naked eye, but the closer the UV range, you can see the signal depending on the molecular composition of these fluorophores, NADH, FAD, and tryptophan that are naturally in your tissue. The excitation is 250-450 nm range. It’s a very weak signal, the fluorescent signal, so you need a good detector to pick it up. Most of the fluorophores that are naturally found are in this table, as I said before. We can use these things to quantify the disease.

In the scattering or the diffuse reflectance basically what happens is the incident light comes in and they get scattered at the gland cell nuclear boundaries. It’s a very strong signal and the same wavelength as the incident light. There’s no change in the wavelength. It depends on the tissue morphology architecture. Consequently, we can actually use this to diagnose not just benign versus malignant disease, but also the histopathological grade such as the Gleason grading. You can use a standard optical probe with fiber optics arranged in certain geometries to send the photons to the tissue and then read it back. Then you quantify the signal and you can use that information to figure out what the disease status is.

We actually ran some cell line data to figure out whether these fluorophores are in there. We ran on the DU 145 and normal cells. You can see the color of the tryptophan and the NADH peaks are there. The collagen peak is not there because it’s a structural protein. It’s not in the cell lines. This is what you see in the fluorescent signal of prostate tissue, actual tissue or – – specimens due to radius excitation starting from 280 nm all the way up to 450. You can clearly see that the tryptophan, the collagen, and NADH peaks are there. The scattering signal is something like this, the absorption feature, slope variations, etc.

If you want to harness the potential of this technology, you need to design a biopsy needle that has the optical probe at the tip, and to do that, at the same time you need to cut a piece of core. If we want to bring the fiber from the back end all the way to the front then that is the specimen notch which you have to bring this fiber underneath. What happens is when you try to do this the question becomes the structural integrity of the needle? When you are filing this needle to cut a core, there are going to be certain g-forces something like 500 or 600 newtons. We actually handed over this project to a group of senior mechanical engineering students as their senior design project. They successfully designed these things for us. They were able to come up with a design where they can actually bring the fiber all the way up to the front tip of the needle underneath that specimen notch so we can actually measure something, cut the same core, same piece of tissue you measured, and then histopathologically – – with that. So that’s the key component here.

There are a number of configurations we have studied in the past. The center fiber is always used for capturing the signal. The surrounding fiber is for excitation. Basically what we have now is this kind of a design and we have been using the BARD Magnum Gun, and this is the biopsy needle. It has the sensor at the tip, and here you have fiber optic cable. You can connect this into a fluorometer and then you can measure the excitation, the emission signal, and classify the disease. So the whole idea is now you can focus your biopsies into these suspicious areas. If the signal says it’s benign, then you probably may not have to take a biopsy.

For the experimental setup, we used a standard laboratory fluorometer. This is from JY-Horiba for our initial ex vivo studies. We get the prostate minutes after Dr. Crawford does the surgeries. The main reason is these fluorophores begin to decay once they are out of the body. You have a window of 90 minutes to capture all this data, and afterwards you don’t do anything. What we do is we take the prostate, insert the needle, and capture the spectra, and then cut a piece of tissue. As I showed before, you can very clearly see there are tryptophan, collagen, and NADH peaks. The excitation for tryptophan is about 290. The imaging peak is around 340. For collagen it is about 340. The imaging peak is about 400, etc. These fluorophores are naturally there.

The challenge is how do you separate benign versus malignant disease. For the diffuse reflectance, again, here are some curves. You can clearly see some absorption features and there are also slope variations. Again, if you can quantify this data you can come up with the disease status. As I said, the biggest obstacle for this type of study is how do you histopathologically classify what you measured? You can do all this work, but at the end of the day the pathologist has to say, “Okay this is cancer. This is not.” How do you correlate this? That’s what we have done for our biopsy needle.

We have found that there is a window. It’s about half a millimeter wide located about 1.7 mm from the core of the biopsy. What we do is when we take these biopsy cores, we ink one in so we know which end we measured. Then the pathologist will classify this based on all the standard classifications, cancer if it is cancer, the Gleason grades are given, atrophy, stroma, BPH nodules. Everything goes into a chart like that. Then what you do is you correlate this data with the spectral data. This is a very straightforward algorithm. You take the raw data and you bring also the histopathological data. You do a background subtraction, and we look for signal-to-noise ratio. If the signal-to-noise ratio is less than six, we discard those signals.

We do some data manipulation. We run something called partial least square analysis. This is similar to principal component analogy, which is a data reduction scheme. You have spectral data. There’s a lot of data so I need to find what we call the features. We can do what we call partial least square analysis where it will compare the X and the Y. X in this case is your spectral data, Y is the histopathologic data. We try to look at the best features that explain differences in the Y data set. Then we use a support vector machine classification for benign versus malignant. The validation is leave-one-out cross validation technique. These are some figures of the curves we have, the fluorescent signals we have captured for the benign tissue and for the malignant tissue. One of the things you have to understand is the absolute fluorescence levels are not really useful because your fluorescence level can change from patient to patient. Because of that, we normalize the spectrum and then you look at the variations. That’s something you have to keep in mind.

We have captured data from 20 prostates, and there are 187 cores, 109 malignant, 78 benign. Out of the 78 malignant, 49 were scored Gleason 6. The other 29 were greater than or equal to Gleason 7. We calculated the two positives, two negatives, false positives, false negatives, then from those you can determine the performance sensitivity specificity data. This is what it looks like. Some of the partial least square components. Depending on how many components you bring in, you can see the accuracy get much better. You can use certain thresholds to calculate your false positives and false negatives. You can also do these ROC curves. Here you can see if you use the 290 excitation, which is a tryptophan signal, seems to be the best. The 330 and 350 which are the collagen and the NADH signals have a little less area under the curve. Also, if you keep adding more components, you can see the area under the curve will also increase. So there are a number of ways you can manipulate the data set. Here is the experimental data. Fluorescent spectra, the sensitivity is about 82%. Negative comparative value is about 78% and 87%. For the scattering, it is less than the sensitivity of the fluorescent spectra but still got better.

We can combine these two and improve the sensitivity and the negative productive values. This says if you get a signal and say it is benign, that means you have a very high negative productive value. The fluorescent spectrum is probably the best classifier out of the two and has very high sensitivity and NPV for combined classification; the lowest specificity and PPV for the combined classification as well.

Here is the classification between the grades, high grade versus benign. The high grade is Gleason score of 7 or greater. Again, you can see form the benign to high grade very good sensitivity. This is using the diffuse reflectance spectra. High negative productive value, probably the weakest, is the high grade versus low grade. The low grade was benign. That’s probably the weakest classification. Again, high grade versus low grade, you can separate them with a fair degree of accuracy.

With this information, we designed a clinical system for clinical validation. The concept of the system is we cannot use that big laboratory fluorometer in the clinical setting, so we built our own custom fluorometer for this study. We used a whole bunch of manufacturers to manufacturer this things outline. This is our fluorometer. It is a very small box. The laptop runs all the software manufactured by Ocean Optix. There are two light sources, 290 and 340, about 1mW power, calibrated to deliver about 5µW power to tissue. The optical biopsy needle was manufactured by Avantes. We have tested it for the mechanical and structural stability. This is the complete system. This is Dr. Crawford doing this open radical prostatectomy and testing the needle at that point. We did this support vector machine analysis. We found in vivo data to be far superior than our ex vivo data, 84% sensitivity, 97% negative productive value.

You can use this to diagnose the disease with high sensitivity and the high negative productive value will give you an indication of benign tissue so you can focus on areas where you think there are suspicious lesions. The limitation is it is a very small sample size, about 13 patients. There are advantages to the device. One of the bigger advantages is you don’t have to change your existing practice. It kind of fits very nicely and eliminates repeat biopsies. You can eliminate false negative biopsies as well. This is a lot of savings. This is the team on the AMC campus and the Precision Biopsy team who participated.

I have one last question. If there is a commercially available device like this where would you use it? For initial biopsy, for repeat biopsy, for both initial and repeat or other application or none? For both initial and repeat biopsies, thank you.

Crawford ED, Hirano D, Werahera PN, et al. Computer modeling of prostate biopsy: tumor size and location–not clinical significance–determine cancer detection. J Urol. 1998 Apr;159(4):1260-4. http://www.ncbi.nlm.nih.gov/pubmed/9507848

Ramanujam N. Fluorescence spectroscopy of neoplastic and non-neoplastic tissues. Neoplasia. 2000 Jan-Apr;2(1-2):89-117. http://www.ncbi.nlm.nih.gov/pubmed/10933071

Richards-Kortum R, Sevick-Muraca E. Quantitative optical spectroscopy for tissue diagnosis. Annu Rev Phys Chem. 1996;47:555-606. http://www.ncbi.nlm.nih.gov/pubmed/8930102

Utzinger U, Richards-Kortum RR. Fiber optic probes for biomedical optical spectroscopy. J Biomed Opt. 2003 Jan;8(1):121-47. http://www.ncbi.nlm.nih.gov/pubmed/12542388