Classifications

• • A—HUMAN NECESSITIES
  • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
  • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
  • A61B5/00—Measuring for diagnostic purposes; Identification of persons
  • A61B5/44—Detecting, measuring or recording for evaluating the integumentary system, e.g. skin, hair or nails
  • A61B5/441—Skin evaluation, e.g. for skin disorder diagnosis
  • A61B5/444—Evaluating skin marks, e.g. mole, nevi, tumour, scar
• • A—HUMAN NECESSITIES
  • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
  • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
  • A61B5/00—Measuring for diagnostic purposes; Identification of persons
  • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
  • A61B5/0075—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy
• • A—HUMAN NECESSITIES
  • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
  • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
  • A61B5/00—Measuring for diagnostic purposes; Identification of persons
  • A61B5/44—Detecting, measuring or recording for evaluating the integumentary system, e.g. skin, hair or nails
  • A61B5/441—Skin evaluation, e.g. for skin disorder diagnosis
  • A61B5/445—Evaluating skin irritation or skin trauma, e.g. rash, eczema, wound, bed sore
• • A—HUMAN NECESSITIES
  • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
  • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
  • A61B5/00—Measuring for diagnostic purposes; Identification of persons
  • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
  • A61B5/7235—Details of waveform analysis
  • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems

Definitions

• the present invention relates generally to the field of spectroscopy.
• the present invention relates to a method for non-invasively diagnosing skin diseases using visible and near-infrared spectroscopy.
• Skin cancer is the most common human cancer. In 1999, it is estimated that there will be 70000 new cases of skin cancer in Canada (Canadian Cancer Statistics: Toronto: National Cancer Institute of Canada, 1999) and more than 1 million new cases in the United States. The clinical diagnosis is often difficult since many benign skin diseases resemble malignancies upon visual examination. As a consequence, histopathological analysis of skin biopsies remains the standard for confirmation of a diagnosis. However, the decision must be made as to which and how many suspicious skin diseases to biopsy.
• Visible/infrared (IR) spectroscopy may be that tool (Jackson et al, 1997, Biophys Chem 68:109- 125).
• the IR spectrum is divided into three regions: near-IR (700-2500 nm), mid-IR (2500-50000 nm) and far-IR (beyond 50000 nm).
• near-IR 700-2500 nm
• mid-IR 2500-50000 nm
• far-IR beyond 50000 nm
• the near-IR region is often sub-divided into the short (680-1100 nm) and long (1100-2500 nm) near-IR wavelengths, based upon the technology required to analyze light in these wavelength regions.
• the heme proteins oxy- and deoxyhemoglobin and myoglobin
• cytochromes dominate the spectra, and their absorptions are indicative of regional blood flow and oxygen consumption.
• Long wavelength near-IR absorptions arise from overtones and combination bands of the molecular vibrations of C-H, N-H and O-H groups. The absorption of near-IR light therefore provides information concerning tissue composition (i.e. lipids, proteins) and oxygen delivery and utilization.
• Visible and near-IR light is brought from a spectrometer to the skin via a fiber optic cable.
• the light penetrates the skin, and water, hemoglobin species, cytochromes, lipids and proteins absorb this light at specific frequencies.
• the remaining light is scattered by the skin, with some light being scattered back to the fiber optic probe.
• the light is collected by the probe and transmitted back to the spectrometer for analysis.
• a plot of the amount of light absorbed at each wavelength (the spectrum) is computed. Measurements are rapid, non-destructive and non-invasive.
• a method of diagnosing skin diseases comprising: providing a patient having a disease; emitting a beam of visible/near-IR light into a portion of the skin afflicted with the skin disease; collecting and analyzing reflected light from the beam, thereby producing a condition spectrum; emitting a beam of visible/near-IR light into a control skin portion of the patient which is not afflicted with the skin disease; collecting and analyzing reflected light from the beam, thereby producing a control spectrum; comparing the control spectrum and the condition spectrum; and identifying the skin disease based on said comparison.
• a method comprising: a) providing a patient having a skin disease; b) emitting a beam of visible/near-IR light into a portion of the skin afflicted with the skin disease; c) collecting and analyzing reflected light from the beam, thereby producing a disease spectrum; d) emitting a beam of visible/near-IR light into a control skin portion of the patient which is not afflicted with the skin disease; e) collecting and analyzing reflected light from the beam, thereby producing a control spectrum; f) performing a biopsy on the portion of the skin afflicted with the skin disease; g) classifying the skin disease based on the biopsy; h) assigning the control spectrum and the disease spectrum to a skin disease group based on the classification; and i) creating a database by repeating steps (a) to (h).
• Figure 2 shows paired t-test results comparing normal and skin lesion near-IR spectra.
• the mean normalized spectra blue and red traces
• the optical density scale refers to the spectra
• the p-value scales correspond to the p-plot traces.
• Figure 3 shows the difference visible/near-IR spectra from skin lesions. Difference spectra were obtained by subtracting each lesion-normal pairing for each group shown in Fig 3. Actinic keratoses (blue), BCC (red), actinic lentigines (green), dysplastic nevi (black), banal nevi (pink) and seborrheic keratoses (brown) are shown. The areas used for analysis of variance (ANOVA) are shaded over the spectra.
• Figure 4 shows optimal classification regions of visible/near-IR spectra from skin lesions. Class average spectra are shown with the regions for optimal classification (GA-ORS) indicated in the darkly shaded regions and the regions that were significant by ANOVA indicated in the lightly shaded regions. Three optimal regions were selected for dysplastic vs. banal nevi (a), five regions for actinic keratoses vs. actinic lentigines (b), five regions for actinic keratoses vs. seborrheic keratoses (c) and four regions for BCC vs. seborrheic keratoses (d). No regions were significant by ANOVA for b and c.
• a "skin condition” is a dermatological disorder that manifests as a rash, irritation or dry skin. Examples of skin conditions are psoriasis, hives, eczema, etc.
• a “skin lesion” is a circumscribed abnormal area of the skin such as a tumor, nodule or papule.
• a "skin disease” is any abnormal area of the skin caused by disease. Skin diseases include both skin conditions and skin lesions (but not injuries due to external insult such as cuts and burns).
• Actinic keratoses are reddish, rough areas of damaged skin which are considered pre-malignant. A small percentage of these lesions develop into the malignant tumor, squamous cell carcinoma.
• Base cell carcinoma or BCC refers to a slow-growing malignant epithelial neoplasm. This type of cancer in usually “cured” by surgical removal if caught early.
• Actinic lentigines are small benign pigmented lesions often referred to as age or liver spots.
• Plastic nevi refer to atypical moles which are considered to be pre-malignant or at greater risk of becoming malignant.
• “Seborrheic keratoses” are common light brown to black skin growths that are benign.
• Boal or benign nevi are common benign moles.
• visible/near-IR spectra were recorded for a number of patients having skin lesions, as described below.
• a spectrum was taken of an unaffected skin portion as a control from each patient.
• a biopsy was also performed on the skin lesion and the results of the biopsy were used to assign the skin lesion to a specific category.
• the disease spectra and the control spectra were then compared using statistical analysis as described below to detect wavelength regions of significant difference between the control spectra and the lesion spectra. These results were then grouped by skin lesion category based on the biopsy results.
• the grouped spectra showed characteristic patterns in the differential spectra over a specific set of wavelengths. As a consequence, these differences can be used to identify or diagnose a skin disease by comparing the visible/near-IR spectrum of a control region to a spectrum taken of the region of interest.
• the skin disease is diagnosed by emitting a beam of visible/near-IR light into a portion of the skin afflicted with the skin disease, and collecting and analyzing reflected light from the beam, thereby producing a spectrum of the diseased skin portion. The process is repeated for an unaffected region of skin, thereby providing a control spectrum. The control spectrum and the disease spectrum are then compared and the skin disease is identified based on the comparison.
• the skin disease is selected from the group consisting of dysplastic melanocytic nevi; banal nevi; lentigines; actinic keratoses; seborrheic keratoses; basal cell carcinoma; and malignant melanoma.
• the control spectrum and the disease spectrum may be compared at wavelengths corresponding to visible/near-IR absorption by oxyhemoglobin, deoxyhemoglobin, water, proteins, lipids or combinations thereof.
• the wavelengths may be selected from the group consisting of: 518-598 nm; 618-698 nm; 718-798 nm; 918-998 nm; 1158-1238 nm; 1418-1498 nm; 1718-1798 nm; and combinations thereof.
• spectra are taken of affected and control regions from several patients. A biopsy is then performed on each of the affected region, which is then used to positively identify the skin condition. The spectra are grouped according to skin condition, thereby forming a database.
• control spectra and the disease spectra in each skin disease group in the database are then reduced to diagnostic wavelengths using a region selection algorithm. This algorithm is then used to analyze spectra from other skin portions so that the disease afflicting the skin portion can be identified based solely on the spectrum, without performing a biopsy.
• Spectra were recorded in the 400-2500 nm range in 2 nm steps using a commercial spectrometer (Foss NIRSystems Model 6500) equipped with a bifurcated visible/near-IR fiber optic probe with a 7 mm active area.
• Each reflectance spectrum was collected with a 10 nm slit width, and consisted of 32 scans, which were co-added to improve signal to noise.
• the subject's skin and the end of the probe were cleansed with 70% alcohol.
• the fiber optic probe was then positioned 0.5 mm from the measurement site by measuring with a micrometer. For all 195 cases, three (3) visible/near-IR spectra were taken from: 1) the lesion and 2) an area of normal appearing skin (the control site). Acquisition of each spectrum took 40 seconds.
• spectra were grouped into one of six lesion categories: 1) actinic keratoses (33 cases, 99 spectra), 2) BCC (32 cases, 96 spectra), 3) dysplastic melanocytic nevi (13 cases, 39 spectra), 4) actinic lentigines (12 cases, 36 spectra), 5) banal common acquired nevi (22 cases, 19 intradermal and 3 compound nevi, 66 spectra) and 6) seborrheic keratoses (18 cases, 54 spectra). A total of 130 cases were thus included in the data set. The remaining 65 cases either did not fit into one of the above categories or the patient declined to have a biopsy after the measurements.
• the histopathology was the "gold standard"
• control spectra were found to lie within 2 standard deviations of the mean spectrum. Once again, control spectra that lay outside 2 standard deviations from the mean were associated with patient movement. Control spectra for each control site were then averaged, resulting in 130 control spectra.
• LDA returns a value ranging between 0 (not belonging) and 1 (belonging) to each spectrum in a data set, indicating the membership in each class.
• the values returned provide an indication of the likelihood of a spectrum belonging to each class.
• Each spectrum is then allocated to the class to which it most belongs.
• EXAMPLE IV - RESULTS The mean control (i.e. from normal skin) visible/near-IR spectrum is shown in Fig 1. Spectra are plotted showing the amount of light absorbed by the skin at each wavelength between 400-1840 nm. Each peak in the spectrum can be assigned to a specific compound found in the skin. Visually, strong absorption bands arising from O-H groups of water dominate the spectrum. However, much information is present in the weaker spectral features. For instance, the relatively strong absorption feature at -550 nm arises from hemoglobin species and provides information relating to the oxygenation status of tissues.
• tissue oxygenation can be obtained from analysis of a weak absorption feature at 760 nm, arising from deoxyhemoglobin (Stranc et al, 1998, Br J Plast Surg 51:210-217).
• Compositional information can be obtained from an analysis of two absorption bands between 1700-1800 nm associated with C-H groups of skin lipids.
• a series of weak absorption bands arising from protein N-H groups is found in close proximity (usually overlapped by) the strong water absorptions.
• tissue architecture/optical properties can be obtained from the spectra. Changes in tissue architecture/optical properties may affect the basic nature of the interaction of light with the tissue.
• changes in the character of the epidermis may result in more scattering of light from the surface, reducing penetration of light into the skin in a wavelength dependant manner.
• different tumor densities i.e. nodular vs. diffuse
• Such phenomena would be manifest in spectra as changes in the slope of the spectral curves, especially in the 400-780 nm region.
• Fig 2 mean normalized lesion spectra (red traces) and control spectra (blue traces) are shown overlaid on corresponding p-plots (black traces). Several areas of the resulting p-plot contained contiguous regions of statistically significant p-values (p ⁇ 0.05). Each lesion-normal comparison exhibited a slightly different p-plot, and therefore, a distinct pattern of significance.
• Spectra from dysplastic nevi were significantly different from actinic keratoses, BCC, lentigines, banal nevi and seborrheic keratoses in a number of spectral regions.
• BCC spectra were significantly different from banal nevi and seborrheic keratoses in three spectral regions, and seborrheic keratoses were different from lentigines in one spectral region.
• Two class LDAs were performed on the following comparisons: 1 ) dysplastic vs. banal nevi, 2) dysplastic nevi vs. lentigines, 3) actinic keratoses vs. lentigines, 4) actinic keratoses vs. seborrheic keratoses, 5) BCC vs. seborrheic keratoses, 6) BCC vs. banal nevi and 7) dysplastic nevi vs. seborrheic keratoses.
• optimal regions were identified by the GA-ORS algorithm.
• Figure 4 shows the optimal regions for comparisons 1 , 3, 4 and 5.
• LDA resulted in an overall accuracy of 97.7-72.4% compared to a clinical accuracy (by visual examination) of 100-78.0% and are shown in Table II.
• the numbers in rows represent the histopathological classification, while results in columns represent the calculated classification.
• the visible/near-IR spectra of skin presented here exhibit strong absorption bands from water and a number of weak, but consistent, absorption bands arising from oxy- and deoxy-hemoglobin, lipids and proteins.
• visual examination of spectra did not show distinct differences in these spectral features that could be used to distinguish between spectra of skin diseases and healthy skin.
• Univariate statistics were therefore applied in order to determine whether differences existed between skin lesions and healthy skin.
• multivariate statistics were performed in an attempt to objectively classify spectra.
• the region 718-798 nm contains the absorption of deoxyhemoglobin, while the region 918-998 nm contains a broad absorption associated with oxyhemoglobin.
• the regions 1158-1238 nm and 1418- 1498 nm contain significant absorption bands from water, and possibly some contribution from protein N-H groups. Thus, it appears as if changes in the amount or structure of water in tissues occur between some types of lesion and control tissues.
• spectral bands attributed primarily to C-H groups of skin lipids populate the region 1718-1798 nm. Significant differences between spectra in this region may imply differences in the amount or structure of skin lipids.
• GA-LDA genetic algorithm guided linear discriminant analysis
• the trained LDA algorithm can then be applied to unknown spectra, and the unknown spectra are partitioned into one of the clinical groupings based upon the spectral pattern found.
• the advantage of LDA is that a combination of spectral regions (which perhaps on their own do not contain sufficient information to allow diagnosis), rather than individual regions, are used to achieve a diagnosis.
• a prescribed accuracy typically > 90%
• the linear discriminant analysis program takes the regions selected by the algorithm and identifies the hyperplane that optimally separates the sets of points corresponding to the spectral classes of interest.
• class assignment of any given spectrum involves computing its distance from all class centroids (i.e. the representative class average spectrum) and allocating it to the class whose centroid is nearest.
• class centroids i.e. the representative class average spectrum
• a value ranging between 0 (not belonging) and 1 (belonging) is given, indicating the membership in each class, with the sum of the membership values for all classes being unity. The value returned therefore provides an indication of the likelihood of the spectrum belonging to each class.
• GA-LDA was applied to difference spectra from benign and premalignant/ malignant lesion groups. Some of the more difficult visual diagnoses were successfully distinguished. All LDA comparisons save one resulted in an accuracy rate greater than 80%. Although the clinical (visual) diagnostic accuracy rate in this particular study was high (greater than 78%), other studies report clinical diagnostic accuracy rates of 42-65% (Pichter et al, 1991 , Br J Dermatol 125 (Suppl 38):93-97; Hallock and Lute, 1998, Plast Reconstr Surg 101 :1255-1261). The LDA results presented here compare favorably with such studies.
• Spectral regions that contained diagnostic information were not the same as those identified by ANOVA, perhaps reflecting the fact that LDA uses combinations of regions (each of which on it's own may not show significant differences between classes) to enable diagnosis.
• many spectral regions identified by GA-LDA suggest essentially the same biochemical basis for distinguishing between classes as by ANOVA. For example regions around 760 nm (deoxyhemoglobin), 900 nm (oxyhemoglobin) and 1200 nm (water) allowed discrimination between actinic keratoses and actinic lentigines.
• regions around 760 nm (deoxyhemoglobin), 900 nm (oxyhemoglobin) and 1200 nm (water) allowed discrimination between actinic keratoses and actinic lentigines.
• the biophysical basis underlying the diagnostic regions remains unclear.
• dysplastic nevi exhibited a highly significant difference (p ⁇ 0.001 ) from almost all other lesion groups across most of the regions tested by ANOVA.
• classification between dysplastic and banal nevi had the highest accuracy of all classifications (97.7%), with classification between dysplastic nevi and lentigines close behind (92%).
• the accurate and early diagnosis of dysplastic nevi is a significant development in the recent emphasis placed on melanoma detection.
• Visible/near-IR spectroscopy could form the basis of a clinical method to diagnose skin diseases. It is rapid (i.e. acquisition time of minutes), simple to perform and non-invasive. Measurements are accurate and reproducible. Collection of spectra causes little or no patient discomfort, does not alter the basic physiology of the skin, poses no hazard to the patient and does not interfere with any other standard clinical diagnostic practices. The test could be performed by a non-specialist and, therefore, might be a useful tool for pre-screening skin diseases.

Landscapes

• Health & Medical Sciences (AREA)
• Life Sciences & Earth Sciences (AREA)
• Physics & Mathematics (AREA)
• Heart & Thoracic Surgery (AREA)
• Molecular Biology (AREA)
• Pathology (AREA)
• Engineering & Computer Science (AREA)
• Biomedical Technology (AREA)
• Veterinary Medicine (AREA)
• Medical Informatics (AREA)
• Biophysics (AREA)
• Surgery (AREA)
• Animal Behavior & Ethology (AREA)
• General Health & Medical Sciences (AREA)
• Public Health (AREA)
• Dermatology (AREA)
• Spectroscopy & Molecular Physics (AREA)
• Investigating Or Analysing Materials By Optical Means (AREA)
• Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Applications Claiming Priority (3)

Publications (1)

Family

ID=22565569

Family Applications (1)

Country Status (4)

Families Citing this family (6)

Family Cites Families (2)

• 2000
  • 2000-10-05 WO PCT/CA2000/001187 patent/WO2001024699A2/en not_active Ceased
  • 2000-10-05 CA CA2396883A patent/CA2396883C/en not_active Expired - Lifetime
  • 2000-10-05 AU AU77662/00A patent/AU782431B2/en not_active Ceased
  • 2000-10-05 EP EP00967472A patent/EP1237479A2/de not_active Withdrawn

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events