Preliminary study for non – invasive optical detection of squamous and basal cell carcinomas

Background The early detection of skin cancer may highly increase the chances of its healing. One of the non-invasive methods of such detection based on the Oblique- Incidence Diffuse Reflectance (OIDR) measurements of the reflected diode laser light from the skin. In this research we designed and implemented the OIDR reflectometry measuring system with a 650 nm diode laser source to aid physicians in diagnosing both squamous cell carcinomas (SCC) and basal cell carcinomas(BCC). Method The laser is delivered obliquely to the skin surface by an optical fiber fitted through a tube holder of CCD camera. The diffused reflected laser light from the skin is captured by the CCD camera and sent to a computer, which is supplied by a specially prepared Matlab program to analyze these images in order to decide in a time whether the lesion is malignant or benign. Fifty cases were diagnosed under supervision of the consultant section of The Governmental Specialized Marjan Teaching Hospital – MOH – Iraq. Result The fifty diagnosed cases by this technique, the results were 90% accurate. Conclusion The method of laser oblique-incidence diffuse reflectance (OIDR) combined with using the developed algorithms that have high classification rates may prove useful in the clinic as the process is fast, noninvasive and accurate.


Introduction
Skin cancer is the most common form of cancers with increasing rate per year especially in fair skin population. Non-melanoma skin cancers account for about half of all cancers and include basal cell carcinomas (BCC) and squamous cell carcinomas (SCC) [1]. "Current diagnostic methods for skin cancers rely on physical examination of lesions in conjunction with skin biopsy, which involves the removal of tissue samples from the body for examination. Biopsy of large lesions often requires substantial tissue removal. Though this protocol for skin lesion diagnosis has been accepted as the golden standard, it is subjective, invasive, time-consuming and painful. Laboratory results for the determination of histopathology of a suspected tumor may generally take several days. Since suspicious areas are identified by visual inspection alone, there are a significant number of false positives that undergo biopsy. Conversely, many malignant lesions can also be overlooked. There is an urgent need for objective criteria that would aid the clinician in evaluating whether biopsy is required" [2]. Now a days there is a growing demand for accurate and fast models to predict the light distribution in biological tissues to deduce their optical properties from the measurable quantities [3]. One of the measurable quantities is the diffuse reflectance, it is a function of the distance between the observation point and the incident point of a laser beam. The diffuse reflectance is defined as the photon probability of re-emission from inside a semiinfinite turbid medium per unit surface area (skin tissue). Measurements of the diffuse reflectance can be used to determine the optical properties of tissue non-invasively [3].
Biological scatterers are primarily cell nuclei and mitochondria, with diameters ranging from 1 μm to 8 μm. As the laser light wavelength is smaller than these scatterers, therefore the light interaction can be predicted by Mie scattering theory, which is an exact analytical solution of Maxwell's electromagnetic field equations, but when the scattering particles are much smaller than the wavelength, the light interaction can be predicted by Raleigh scattering theory, which is a limiting case of Mie theory. Scattering coefficient is defined as the probability of photon scattering per unit infinitesimal path length [4].
In this paper we present a design and implementation of a non-invasive, painless and fast method to deduce the optical properties of skin cancerous suspicion lesions based on the application of the oblique incidence diffused reflectance reflectometry (OIDR) as originally conceived by Wang and Jacques [5,6].
The images of the diffused reflectance for both lesion and healthy adjacent skins of the same patient are captured by a CCD camera, these images are then analyzed and processed by a specially written Matlab program v.10 to perform a logical prediction (diagnosis) for the examined lesion for squamous and basal cancerous cells.

Materials and methods
Patients studied SCC and BCC are types of non-melanoma skin cancer that chosen by Research Section of Medical Engineering / College of Engineering / Al-Nahrain University.
Fifty patients are selected under supervision of Dermatology Consultant Section of The Governmental Marjan Teaching Hospital / MOH -Iraq. Dermatologist identify suspicious skin lesion that were going to be biopsied for routine care. Patients were asked to participate in the study and sign on informed consent approved by the Babylon Health Directorate / MOH -Iraq.
The needed data from each patient were collected, analyzed and recorded before physicians removed the lesion and sent it for biopsy. Histopathological diagnoses was performed by the specialized laboratory of Marjan Teaching Hospital and reported within 6 to 7 days to compare with our recorded logic predicted result.

Oblique incidence diffused reflectance method
The significant changes that happen in malignant cells make it optically differentiable from benign cells due to the enlargement in their cell nuclei and mitochondria sizes, it is important to know that the nuclei and mitochondria are the major scatterers in the cells, therefore the enlargement of their size considered as the important indicator to the presence of cancer cells that cause increase in light scattering [4,7].
When a light enters a semi-infinite tissue, it will generally scatter many times before either being absorbed or escaping the tissue surface at a point other than its point of entry. The multiple scattered light that escapes is called diffuse reflectance [8,9], as seen in Figure 1.
Because it is easier to model isotropic scattering than anisotropic scattering, the reduced or transport scattering coefficient μ s ' is introduced as the equivalent isotropic scattering coefficient of an anisotropically scattering medium. μ s '=μ s (1g), where μ s is the scattering coefficient and g is the average cosine of the scattering angle [8,11]: A sketch for the laser oblique incidence diffuse reflectance pattern from a semi-infinite turbid medium like the biological tissue is shown in Figure 2.
The spatial distribution of diffuse reflectance of an oblique incident laser beam from a semi-infinite turbid medium like the biological tissue has been modeled according to Wang and Jacques by two isotropic point sources; one positive source located below the tissue surface and one negative image source above the tissue surface, as shown in Figure 3. The positive source is buried at distance (d s ) from the point of laser incidence on the skin, this distance is considered practically to be three times greater than the diffusion coefficient "D" as in the mean free path Eq.(1) [6,11,12]: The modified dipole source diffusion theory model gives diffuse reflectance at the skin boundary R(x), by using Eq.(2) which can be scaled to fit a relative reflectance profile that is in absolute units [6].
Where ; Figure 1 Light interaction in a scattering and absorbing media [10].
α i : is the angle between the incident laser beam and the normal line on the tissue surface. α t : is the angle of light transmission into the tissue, could be calculated according to Snell's law which is used to measure the new optical path where the isotropic positive point is locate. as seen in Figure 3. r : is the distance between the normal line between the positive and negative point and the observation point.
x : is the distance between the point of observation and the point of light incidence (origin point). ρ 1 , ρ 2 : are the distances from the two point sources to the point of interest. z b : is the distance between the virtual boundary and the surface of the tissue. A : is the parameter related to the internal reflection which can be calculated using either Fresnel reflection coefficients or using empirical variable r i and relative reflection coefficient n rel of the tissue ambient (air) interface as following [8]: Figure 2 Single wavelength oblique incidence diffuse reflectance pattern [10].
A : is unity for a matched boundary [8].
Δz : is the depth of the positive point source from the surface of skin.
Δx : is the distance shift between the point of laser incidence and the center of the most symmetrical circle, as seen in Figures 2, 3 and 4.
Ones the distance shift (Δx) was found, the diffusion coefficient D could be calculated from Eq.(8) [6]: Figure 4 Oblique incidence diffuse reflectance curve along the x direction [10].
μ eff is the effective attenuation coefficient [6]: The μ eff value could be found by using a least-square fitting to Eq.(2). Now it is possible to find the skin optical properties μ s ' and μ a from Eqs.(10 and 11) as following [6] : Experimental work

Experimental setup
The oblique incidence diffused reflectance reflectometry (OIDR) system that is designed in this work to measure the skin optical properties is sketched in Figure 5. It includes the following components:  The design of CCD camera holder is shown in Figure 6. The virtual center line (axis) of CCD camera was fitted particularly to be on the center of the horizontal plane, exactly at intersecting point with the needle tip.
The CCD camera holder was painted by black color to reduce the effect of the outside light. The clinic lights were turned off during the examination. All the apparatus were placed on a small portable hospital cart as shown in Figure 7 to move it easily in the patient examination room.

Work procedure
The lesion and adjacent healthy area were identified and marked by the dermatologist visually with aid of special lenses.
Placing the special holder tube, Figure 6, on the lesion (without pressing) then using the "c" character on the computer keyboard five times to capture five diffused reflected profile images for the suspension lesion while continuously running the laser, this procedure takes about three seconds. Then we repeat this procedure for the adjacent healthy area.
By using "c" character ( capture command ) on the keyboard the computer will collect the raw data from the CCD camera and import it directly to Matlab program directory folder for analysis.
The measurements are repeated five times in order to average the optical properties for each of lesion and healthy skin.
The optical examination for skin cancer diagnosis was carried on by the sequence shown in Figure 8. Figure 9A shows a sample of the diffused reflected image of case no. 08 in Table 1, it was a low grade malignancy cancer, its relative diffused reflectance curve resulted by the Matlab program computations is shown in Figure 9B.
For the same patient case no. 08 in Table 1, Figures 9C and 9D show the diffuse reflected image of the healthy adjacent skin area and its relative diffused reflectance curve resulted by the Matlab program computations respectively.

Results and discussion
The results are tabulated for fifty examined cases presented in Table 1. Our experimental non-invasive logical prediction decision, beside the invasive biopsy analysis result is shown in the table as well as the optical parameters of the lesion and the healthy skins for each patient. Figures 10 and 11 show the optical properties of the all fifty cases.
This OIDR setup required at least 8 bit of dynamic range of CCD camera to measure the diffuse reflectance within a few centimeters radius of diffused reflectance image, without bring the CCD camera in saturation state.
The accuracy of Δx shift value measurement was proportional to the resolution of the CCD camera in the setup. In our study, the size of pixel was 0.05 mm/pixel and the CCD camera dynamic range was 16 bit.
The sharpness and Δx measurement accuracy of the diffused reflected images, as were shown in Figures 9A and 9C, increased after painting the camera holder tube black.
The 620 -670 nm visible light range is better for these applications than the UV and IR. A red 650 nm diode laser was selected for these measurements due to its low absorption in the high scattering epidermal tissue, which increased the accuracy of the measured optical properties. While using a green laser of 532 nm wavelength to check this phenomena shows that the reflectance at 532 nm was below the sensitivity of the CCD camera system.
The diagnostic logical decisions in Table 1 was based on the μ s ' threshold value selection beside the increment in the μ a value, Figure 12 shows the decision rule flow chart for the 50 patients. If the difference value between the scattering coefficient of lesion and the scattering coefficient of healthy adjacent skin greater than or equal to 1.8 cm -1 Import the captured lesion and healthy images to the computer Morphological process to extract most symmetrical circle then analysis the extracted circle to find the center Calculate the reflectance at points along the diameter of image , then last square curve fitting procedure to data point w.r.t. image diameter (x-axis [mm]) Fiq. (10) Calculate the value of µ eff by Eq. (2), with error less than 0.0001, then calculate the valueof µ a by Eq(10)and µ s ' by Eq (11) End Calculate the x shifting , then calculate D by using Eq. (8) At point x , calculating the value of 1 and 2 by Phithagors law of triangles ,then determine the fitted R(x) at the same point x logical prediction for the optical properties.
print the lesion prediction status Start Figure 8 Work procedure block diagram. and less than or equal to 4 cm -1 and combined with a small increment in the absorption coefficient of lesion over the absorption coefficient of normal healthy adjacent skin, then this lesion will be considered as a low grade malignancy lesion case. The threshold value (1.8 cm -1 ) was chosen according to case no. (09) in Table 1, which is of a 35 year male with lesion in the cheek. The biopsy result was low grade malignant and the difference between the reduced scattering coefficient of lesion and adjacent healthy skin was 1.8511 cm -1 .
While, the threshold value 4 cm -1 was chosen according to case no. (15) in Table 1, 47 year female with lesion in the cheek, the biopsy result was low grade malignant and the difference between the reduced scattering coefficient of lesion and normal healthy adjacent skin was 3.9608 cm -1 . Therefore, the low grade malignancy threshold considered to be ranged from 1.8 cm -1 to 4 cm -1 .
If the difference value between the scattering coefficient of lesion and the scattering coefficient of healthy adjacent skin less than 1.8 cm -1 and combined with a small increment in the absorption coefficient of lesion over the absorption coefficient of normal healthy adjacent skin, then this lesion will be considered as a benign lesion case.
If the difference value between the scattering coefficient of lesion and the scattering coefficient of healthy adjacent skin greater than 4 cm -1 and combined with a small Figure 9 Diffused reflected images for both lesion and healthy adjacent skin area, with their relative diffuse reflectance curves of case no. 08 in Table 1.   Finally, all patients predicted negative on our test ware diagnosed negative (based on biopsy result), therefore there is no false negative (FN) in our procedure. This was accomplished by choosing the precise (1.8 cm -1 ) threshold value.  Proportion of 9 cases that tested negative (TN) of all the 14 patients that actually are negative (TN+FP) represent the Specificity (TNR) of our prediction which it 64.62% (With higher specificity, fewer suspected lesion patients are labeled as cancerous).
While, the Sensitivity (TPR) of our prediction results is 100% (its represent the probability that our test is positive given that the patient has a cancer), this was accomplished by having no false negative (FN) cases. Table 2 shows the statistical results.
Patients cases no. (13, 16, 18, 29, and 48) in Table 1 were predicted by the Matlab logical prediction as a "low grade malignancy", while the patients histological examination (biopsy result) show negative result for malignancy behavior, these false positive (FP) cases are appeared because the difference value between the reduced scattering coefficient of lesion and normal healthy adjacent skin being close to 1.8 cm -1 . Figure 13 shows the results of Matlab logical prediction with two lines of the scattering threshold that represent sensitivity zone of low grade malignancy, the yellow points represent the low grade malignancy cases, the red points represent the malignancy cases, while the blue points represent the benign lesion cases.
Low grade malignancy decision by our system is a critical decision, because it depend on the experience of the physicians to see the value of the difference between the normal and    Figure 13. When the difference between normal and healthy of μ s ' is close to 4 cm -1 threshold value, the physician must expect the malignancy behavior of the tested lesion and send the patient to perform a histological examination to be sure about the diagnosis. While, if the difference between normal and healthy of μ s ' is close to 1.8 cm -1 threshold value, the physician must expect the benign behavior of the tested lesion. The threshold value used to eliminate the false negative (FN) diagnosis, e.g. if we change the threshold (difference) value 1.8 cm -1 to be 1.9 cm -1 , this will produce a false negative diagnosis in case no. (02).

Conclusions
Only the relative profile of the diffuse reflectance was used for the extraction of optical properties, which made the device insensitive to variations of some system parameters such as intensity of the laser source.
The accuracy of Matlab logical prediction result depends on the image analysis and fitting accuracy to obtain optical properties of lesion and adjacent healthy skins, These results in turns depend on many factors such as the light source wavelength, resolution of CCD camera, and the procedure of estimating the center of the circles and determining the Δx shift, as well as the chosen μ s ' threshold value.
The overall accuracy of a our predictive measurements is 90% which represent the degree of closeness of our predictive measurements to the actual (true) diagnoses of suspected lesion, see Table 2.
The common site of SCC and BCC is on the face, therefore it was important to measure the accuracy of the method with respect to the lesion site, the accuracy of the measurements was 91.6% for all lesions that are on the cheek, Table 3. The mean value of the reduced scattering coefficient for the healthy cheek skin is 2.8 cm -1 , that have the statistical variance of 0.1842, and the standard deviation of 0.4292 cm -1 .

Competing interests
The authors declare that they have no competing interests.
Authors' contributions AMA obtained funding for the study, design and implementation the OIDR setup, performing the experimental work, programming the engineering and statistical calculations by Matlab, interpretation of analyzed data, editing, drafting and finalizing the manuscript. MSD supervised the design of OIDR setup, advice the engineering and scientific bases of the research, interpretation of analyzed data, coordinated all the work, drafting and finalizing the manuscript. MKT prepared and selected the patients who are under our study, coordinated the hospital acceptance for the patients