Analysis of skin lesions using laminar optical tomography

Abstract

Evaluation of suspicious skin lesions by dermatologists is usually accomplished using white light examination and direct punch or surgical biopsy. However, these techniques can be imprecise for estimating a lesion's margin or level of dermal invasion when planning surgical resection. Laminar optical tomography (LOT) is an imaging technique capable of acquiring depth-sensitive information within scattering tissues. Here, we explore whether LOT data can be used to predict the depth and thickness of pigmented lesions using a range of simulations and phantom models. We then compare these results to LOT data acquired on normal and malignant skin lesions in vivo.

Keywords: (170.3880) Medical and biological imaging; (170.3890) Medical optics instrumentation.

Fig. 1

Schematic diagram of Laminar Optical Tomography system. lower inset: illustration of offset detector geometry and average photon pathways in tissue (blue arcs). Note that wider detector offsets yield deeper average photon migration pathways. Upper inset: Monte Carlo-derived depth sensitivity of LOT measurements at four source-detector offset positions for optical properties approximating skin (normalized to 1 for each offset).

Fig. 2

Phantom and modeling results. (a) White light photograph of depth-varying optical phantom with the depth of each inclusion indicated. (b) Raw LOT phantom data inclusions of different depths displayed as differential contrast (ΔS/So). (c) and (f) Illustrations of depth-varying and thickness varying optical phantoms, respectively. (d) and (g) Differential contrast measured data (solid lines) and Monte Carlo simulation predictions (dashed lines) for lesions of different depths and thicknesses. Error bars show standard deviations of 15 repeated scans. Power law fits to experimental LOT data are also shown. RMS errors of these fits range from 0.0256 to 0.0666. (e) and (h) plots showing Monte Carlo-derived isolines of power law fit parameters ‘a’ and ‘b’ (y = ax^b), for various thickness and depths of inclusion. Triangles show results of the power law fits for phantom measurements from (d) and (g) respectively and agree well with phantom parameters. We note that compression of ‘thickness’ isolines at higher “a” values is a result of ΔS/So values approaching 1, corresponding to very little light reaching detectors. This is an intrinsic detection limit of LOT, since if light cannot reach the detector, it cannot distinguish between lesions of increasing thickness beyond that threshold.

Fig. 3

Effects of background scattering properties on fit parameters. Filled circle represents the predicted fit parameters for an inclusion at the specified depth and thickness in (left) our phantom model (μ_a = 0.2 mm⁻¹, μ_s = 4.0 mm⁻¹, and g = 0.8) and (right) in our skin-like model (μ_a = 0.2 mm⁻¹, μ_s = 45 mm⁻¹ (100 micron thick epidermis) and μ_s = 30 mm⁻¹ (dermis) and g = 0.8). The colored triangles represent fit results when the background scattering coefficient is increased by 10% (upward pointing triangles) or decreased by 10% (downward pointing triangles). While the absolute prediction for each location changes somewhat, the trends related to relative depth and thickness are retained.

Fig. 4

LOT measurements of benign melanocytic nevi. (a),(d) White light images of Compound and Junctional nevi respectively. (b),(e) Raw LOT image (source-detector separation = 0.25mm) showing reference region (white box) and measurement region (red or blue box) for differential contrast calculation. (c),(f) Representative hematoxylin and eosin (H&E) histopathology images of junctional and compound nevi (40X) [28]. Scale bars approximately 100 microns. Arrowheads demonstrate location of melanocytes. Note that melanocytes in the compound nevus extend deep into the dermis, while melanocytes in the junctional nevus are limited to a thin region at the dermal – epidermal junction. (g), (h) LOT differential contrast and plot of fit parameters. Error bars show standard deviations of 15 repeated scans. RMS error of fit was 0.0096 and 0.0266 for the compound and junctional cases, respectively. Fit results are consistent with the compound nevus being substantially thicker than the junctional nevus, with both lesions being superficial.

Fig. 5

Clinical squamous cell carcinoma. (a),(b) White light and LOT image at 0.20mm source-detector separation (inset) of squamous cell carcinoma lesion (metric ruler at left for scale). Yellow dashed line indicates approximate location of histopathology section from superior (S) to inferior (I) margins. (c) 532 nm LOT image showing regions of interest selected for differential contrast measurement. White box = reference region, Blue and green boxes = regions of interest. Red box = normal region. (d) Raw LOT frames taken at 532nm and RGB merged data for 5 source-detector (s-d) separations. Anomalous red structures are highlighted by yellow arrows (and black arrows in (b)). Note that the pink circles in wider separations are a specular reflection caused by the variable position of the articulating arm between imaging and calibration scans [21]. (e),(f) Plots of differential contrast and fit parameters of regions in (c). Error bars show standard deviations of 15 repeated scans. RMS error of fit for the red, green, and blue ROIs were 0.0462, 0.0528, and 0.0622, respectively. Note that the blue and green regions appear to be deeper than the red region (which is off scale to the left, implying a very superficial source of contrast). (g) H&E histopathology section from yellow dashed line. Scale bar 200 microns. Red arrowheads show increased subsurface blood accumulation, black arrowheads show adjacent aggregates of dysplastic keratinocytes.