Optical Analysis of Biomedical Specimens

FIG. 1 is a block and schematic diagram illustrating apparatus for carrying out one of the techniques of the invention;

FIG. 2 is a block and schematic diagram for carrying out another of the techniques of the invention;

FIG. 3 is a typical graph of optical density versus wavelength for rat livers with and without cancer;

FIG. 4 is a line-plot of x-values of the correlation equation for rat liver cancer;

FIG. 5 is a typical graph of optical density versus wavelength for mice with and without cancer;

FIG. 6 is a line-plot of x-values of the correlation equation for cancerous mice;

FIG. 7 is a typical graph of optical density versus wavelength for contaminated and control serum; and

FIG. 8 is a typical graph of dT/T versus wavelength for a nonmalignant cervical scraping.

The optical instrumentation 10 in FIGS. 1 and 2 includes an infrared lamp 12 which directs wideband light through a lens 14 to illuminate a focused area on a sample drawer 16. The drawer 16 is adapted to carry a specimen 18 which may be in liquid, solid or gaseous form. The sample drawer 16 should be apertured so as to present no obstacle but the specimen to light above and below the focused area. A multiple filter assembly 20 in the form of a paddlewheel is mounted for rotation about an axis a perpendicular to and spaced from the light path defined between the lamp 12 and the illuminated area on the sample drawer 16. The paddlewheel is typically driven at about 600r.p.m. In the preferred embodiment, three narrow bandwidth plate-shaped optical interference filters 22 are mounted on a hexagonal axle 24. The respective planes of the filters 22 are equally separated angularly by about 120.degree.. The filter extend radially from the axle in planes which intersect approximately in a line conincident with the axis of rotation a. The filters 22 extend equal radial distances from the axis a. The configuration of the filters 22 on the axle 40 thus resembles the letter .UPSILON.in cross-section. Opaque vanes can be attached to the ends of the filter to intermittently abstruct the light path.

In FIG. 1, the pulse point output 26 of the paddlewheel is passed through a counter circuit 30 which can be set as discussed in U.S. Pat. No. 3,861,788 to provide individual control signals corresponding to selected angles in the rotation of the paddlewheel and thus to specific wavelengths of interest. The output of the photodetector 28 is passed to a low noise sensor amplifier 32, which is implemented as shown in U.S. Pat. No. 3,861,788. The output of the sensor amplifier 32 is fed via a logarhythmic amplifier 34 to four sample and hold circuits 36, 38, 40 and 42 gated respectively by four outputs of the settable counter circuit 30. Each sample and hold circuit is gated to store a value of optical density at a selected wavelength. An analog computer 43 receives the output of the sample and hold circuits 36, 38, 40 and 42 to multiply the individual optical densities by constants and to sum the resulting products according to a programmed expression. The value of the resulting expression relative to a predetermined level is related to a property of the specimen.

If linear expressions involving .DELTA.OD instead of OD were desired, for example, for eight wavelengths, the number of sample and hold circuits could be doubled or the arrangement using a differential amplifier and two stages of sample and hold circuitry could be used as shown in U.S. Pat. No. 3,861,788.

In FIG. 2, the output of the low noise sensor amplifier 32 receiving the photodetector output 28, instead of going to the subsequent analog circuitry in FIG. 1, is fed to analog to digital converter 44. The multibit digital output of the converter 44, representative of transmissivity at the various wavelengths passed by the paddlewheel 20 is fed to the parallel data input of a digital computer 46. The data input of the digital computer 46 is gated by means of the pulse point output 26 which acts as a clock signal. The circuitry in FIG. 2 is inherently superior to that in FIG. 1 for one primary reason: storage capacity.

If the output of the sensor amplifier is digitally encoded into 12 parallel bit channels, and of the 1,000 pulse points per revolution, there are only 100 nonredundant useful points per filter for a total of 300 pulse points which correspond to wavelengths, the amount of digital memory necessary to store values of transmissivity for every one of the 300 wavelengths is only 300 .times. 12 = 3,600 bits. To store 300 more points for absolute transmissivity without the specimen, the storage capacity should be doubled to 7200 bits. In comparison, the equivalent of the simultaneous storage capacity of the analog equipment in FIG. 1 is only 48 bits. Of course, with the digital embodiment of FIG. 2 in a commercially practical system, one would provide only as much storage as needed and program the input to the digital computer 46 to transfer the output of the converter 44 to memory only for selected pulse points. For example, if only 80 pulse points were needed for computations involving absolute and specimen transmissivity, the amount of storage would be 12 .times. 2 .times. 80 = 1920 bits.

The digital computer 46 should be one programmed or programmable to carry out specified arithmetic operations using the stored values of transmissivity as well as a plurality of constant factors which are fed into the computer to carry out a particular formula.

Transmissivity, T, is defined as ratio of the intensity of transmitted light, I.sub.t to the intensity of incident light, I.sub.i at a given wavelength. Absolute transmissivity, T.sub.abs is defined as the intensity of light transmitted to the photodetector in the absence of a specimen at a given wavelength.

If the test calls for a reflectance measurement rather than transmittance, one or more photodetectors can be arranged above the sample drawer 16 to receive light reflected from the specimen, as shown in FIG, 1 of U.S. Pat. No. 3,861,788.

Optical density at a given wavelength, e.g., 1935 A and the difference between optical densities at two different wavelengths, e.g., 1935 A and 1680 A are defined as follows:

Ti, od.sub.1935 = log (1/t.sub.1935),

where, TI, T.sub.1935 = (I.sub.t /I.sub.i).sub.1935;

thus, TI, OD.sub.1935 - OD.sub.1680 = .DELTA. OD = Log (T.sub.1680 /T.sub.1935) = Log (I.sub.tl680 /I.sub.t1935).

If one is analyzing water alone with a substance of constant transmissivity, the value of .DELTA.OD at the wavelengths in the expression above can be used in the following linear expression to arrive at a moisture percentage measurement:

Ti, % h.sub.2 0 = k.sub.O + k.sub.1 .DELTA.OD,

where k.sub.0 and k.sub.1 are constants.

As discussed in U.S. Pat. No. 3,861,788 where the specimen contains not just water but other substances of varying transmittance, including for example, protein and oil, the simple expression using one value of .DELTA.OD is inaccurate. Each of the substances in the specimen has its own spectral absorption curve and thus the overall transmissivity of any given wavelength is affected by the spectral absorption curve of each substance in the specimen. Thus, it has been found that by using characteristic .DELTA.OD'for the spectral absorption curve associated with each substance of primary interest in the expression for just one substance, the results obtained through use of these percentage-determining equations can be correlated more closely with analytical results. For example, as discussed in U.S. Pat., 3,861,788, for agricultural products a more useful expression for moisture percentage is k.sub.0 = k.sub.1 (.DELTA.OD.sub.w) = k.sub.2 (.DELTA.OD.sub.O) = k.sub.3 (.DELTA.OD.sub.p), where, w, 0 and p refer to the characteristic wavelength pairs for water, oil and protein, respectively, at which optical densities are measured and their respective differences obtained.

The following examples illustrate new biomedical applications of spectral transissivity correlation. In each of these tests the basic parameters were chosen, either optical density or the value of dT/T at a particular wavelength, and the values of the parameters for a plurality of sample specimens under observation were recorded for different wavelengths in a computer memory. After the tests were complete, each specimen was analyzed by conventional laboratory techniques to determine whether or not the specimens had a given abnormality such as malignancy. In the case of the fourth example, using cervical cancer, "maybe"was thrown in as a third level variable. Next the affirmative or negative results of the laboratory tests on each individual sample were entered into a computer for correlation analysis with, for instance 300 values of OD stored for each sample. Next came a correlation analysis which is a conventional technique for arriving at an expression involving a mathematical function of the spectral data for any specimen such that the respective values of that expression for all specimens which are affirmative (e.g., malignant) are higher than the respective values of that expression for all specimens which are negative according to the laboratory results. There are many known techniques for correlating two sets of data like the spectral and laboratory data. In all of the examples which follow, except experiment No. 1the type of correlation used to arrive at the expression known as a correlation equation is multiple linear regression analysis assuming a Gaussian distribution of samples even though the pathology information was only in the form of affirmative and negative conclusions. That is no "degree"or "level"of cancer was stated. It is recognized that a Gaussian assumption is not valid for simple binary data. Nevertheless, this approach appears to work well for selecting the optimum absorption wavelengths as evidenced by the data below.

It is important to recognize that the type of computer on which this correlation analysis is performed is not the type shown in either FIG. 1 or FIG. 2. The apparatus shown in the FIGS. relates to special purpose instrumentation which uses the correlation equation arrived at by means of multiple linear regression analysis performed on another computer with much greater capacity and flexibility.

The following four experiments were run to determine whether useful correlation equations could be produced just for the sample specimens. The subjects of the test were are follows: (1) six rats, some with cancer of the liver; (2) 22mice, some with cancer of the bladder; (3) 20 vials of 25% normal human serum albumin, some with microorganisms; and (4) 22 cervical scrapings, some with malignancy, some without and some questionable. In the first three experiments, the test parameter was OD. That is, optical instrumentation like that shown at 10 in FIGS. 1 and 2 was used to produce a photodetector output to a sensor amplifier whose output was fed to a logarithmic amplifier before being fed to a digital computer. The fourth test on cervical scrapings used the parameter dT/T.sub.abs. This term is the ratio of the derivative of transmissivity of the specimen at a given wavelength to the absolute transmissivity without the specimen at that wavelength. The parameter dT/T.sub.abs is more sensitive than the parameter OD and thus one could expect an improvement in the correlation between the spectral and laboratory data in the first three tests if the parameter dT/T.sub.abs was used instead of OD. The parameter d.sup.2 T/T, that is the second derivative of the specimen transimissivity over the absolute transmissivity, may produce even greater correlation although it is more difficult to instrument. The OD correlation equations generated for the first three experiments could be implemented as shown in FIG. 1. The additional mathematical complexity of the dT/T computation in experiment No. 4 makes analog equipment impractical and necessitates digital instrumentation as shown in FIG. 2.

A comparison of the correlation equation results with those of simple observation is presented in Table 1 below, where N and C mean normal and cancerous, respectively:

Like FIG. 3, the above table indicates that all samples were correctly predicated except for rat liver No. 6. Because there were only six samples, only a primary pair of wavelengths were found. Thus, the above correlation could not take advantage of multiple linear regression of wavelength pairs. It was noted for these rats that the cancerous livers had higher optical absorptions than the normal ones, probably because the tumor had a different refractive index than the liver and thus increased the scattering of light.

It was possible to do the mice tests in vivo because the transmittance typically produced optical densities below 5.0 OD. Each mouse was anesthetized with ethyl ether and then its feet were tied to the corners of the rectangular frame. The animal was positioned so that the collimated monochromator beam, approximately 5 millimeters in size, intersected the bladder region. The optical energy impacting the mouse was measurable in milliwatts.

The large difference in absorption between 5900 A and about 10,000 A in FIG. 5 was a characteristic separation between malignant and nonmalignant for almost all mice tested. In fact, prior to receiving the pathology results, a "blind test" was performed using all the spectrum traces. This was done by sorting all the spectrum curves into two groups: those with high absorption between 5900 A and 10,000 A, and those with lower absorption. This simple visual judgement proved correct on 20 of the 22 mice; the only "error" was for malignant mice 11 and 12. The results of this visual examination of the spectrum traces is summarized in the "visual" column of TAble II below:

In this experiment, the larger number of test specimens enabled use of two pairs of wavelengths in multiple linear regression analysis to correlate the spectral and pathological data.

The mice and rat data raises the following questions:

1. Is the correlation really relating the spectrum data to the presence of malignancy? Or could it be that the spectrum correlation is actually relating the presence of the pesticide residue that was in the diet of the cancerous mice and rats, or perhaps some side effect due to the pesticide

2. It is usually expected that the presence of an additional layer of material --i.e., malignant cells on the gladder --would cause an increase in "light scattering" because of the cahnge in refractive index. An increase in light scattering should cause the optical density to increase. Yet the spectrum traces for the cancerous mice show a decrease in optical density with the presence of a tumor.

Experiment No. 3 demonstrates that light transmission can be used as a high speed sorting means of separating contaminated vials on the production line.

Control and contaminated samples are arbitrarily given the nominal values one and ten respectively. In Table III below, the "computed" column represents the value, X, of the correlation expression for the given sample. The "difference" column indicates how close the computed value came to the nominal value. There is a clear separation (screening limit) between the control versus the contaminated samples; the highest computed X-value for a control sample is 5.996 and the lowest for the contaminated samples is 6.310. This separation appears to be statistically significant.

Experiment No. 4 demonstrates the potential of light transmittance analysis as a rapid screening method for cervical cytology samples used in the detection of uterine cancer. The previously studied samples were assigned nominal values of 1,6 and 10 to indicate, respectively, that they were malignant, questionable or nonmalignant. These laboratory values were then correlated with the spectral data for dT/T.sub.abs by multiple linear regression analysis assuming Gaussian distribution even though the sample population was not necessarily normal. On the computer, the dT/T values are produced by plotting all 300 values of transmittance, T, developing an equation to fit the plotted curve, taking the first derivative of the equation, and computing the value of all dT/T.sub.abs over all 300 discrete wavelengths. As discussed below, this procedure differs from a practical instrumentation approach to deriving the values of dT/T.sub.abs, for instance, at four different wavelengths.

The term dT/T was chosen over OD because it produces more sensitive low noise data analysis. A second derivative term d.sup.2 T/T.sub.abs could also be used for this purpose.

Table IV below summarizes the results of the correlations using the X.sub.1 and X.sub.2 correlation equations.

Using the X.sub.1 correlation data with a screening limit of approximately 8.6, there would be 25% false positives (3 out of 12), all six questionable samples as well as all 4 positive samples would have to be screened as suspected positives.

It may be interesting to note, that Sample 13 which has the strongest false positive is from a patient that had cancer of the endometrium which rarely show up in cervical scraping. Also, Sample 16 of the "questionable" group showed up strongly positive, although no cancerous cells were uncovered in diagnostic cytology. This patient had had radiation treatment for cancer of the cervix. Furthermore, Sample 22, to which the X, correlation equation gave the strongest positive indication is from a patient with untreated cancer.

The X.sub.2 correlation equation was based only on the 16 samples that the clinical analysis indicated were definitive (10 negative samples and 4 positive samples). Thus, the optimum wavelengths and correlation coefficients were derived assuming an arbitrary value of 10.0 for the negatives and 1.0 for the positives.

Using the X.sub.2 equation with the screening limit remaining at 8.6, there would be 33% (4 out of 12) false positives; five out of six questionable samples and all four of the positive samples would have been screened as positive samples. The one questionable sample that would have been sorted as negative, Sample 18, is from a patient that had squamous carcinoma that was treated with radiation and sub-total hysterectomy. This sample had some atypyical cells that might be due to healing.

It is noted that expanding the spectrum down to at least 1000 nm will allow thicker samples and even less instrument noise. Larger sample populations should of course be included and more sophisticated data reduction techniques (non-linear regression; d.sup.2 T, etc.) should be evaluated.

The illustrative correlation equations developed in the foregoing examples are used by the devices shown in FIGS. 1 and 2. In the experiments 1, 2 and 3 using OD expressions, the value of the given correlation expressions are computed on analog equipment since data from only a small number of wavelengths are required. Thus, for example, in testing mice, the analog computer 43 would be programmed to compute the value of the linear expression 239.1 - 57.2 OD.sub.1 = 54.2OD.sub.2 - 42.3OD.sub.3 = 16.0Od.sub.4, the OD values being supplied for a given mouse by the sample and hold circuits 36, 38, 40 and 42. The wavelengths 1, 2, 3 and 4 are selected by setting the counter circuit 30 to provide gating signals to the respective ssmple and hold circuits at the corresponding pulse points produced by the paddlewheel filter assembly 20.

In implementing the dT/T correlation equation of experiment No. 4, the digital system of FIG. 2 provides the needed storage capacity. The simple arithmetic of the linear expression could be carried out easily by an analog computer. However, unlike OD, which is derived in one logarythmic conversion, dT/T.sub.abs requires computation of the slope of the transmissivity curve at each wavelength and thus much more data is required. A preferred approach with limited digital storage is to average the difference between 10 pairs of points on either side of wavelength 1, as illustrated in the following program outline:

Another application of the above described principle is the use of light transmittance directly through fertilized eggs to determine whether the embryo is alive.

The above described embodiments and examples relating to the invention are intended to be illustrative, not restrictive. For example, reflectance can be used instead of transmittance. The scope and spirit of the invention are indicated by the appended claims and all equivalents thereto are intended to be embraced therein.