Quantitative melanoma diagnosis using spectral phasor analysis of hyperspectral imaging from label-free slices

Abstract

Introduction: Melanoma diagnosis traditionally relies on microscopic examination of hematoxylin and eosin (H&E) slides by dermatopathologists to search for specific architectural and cytological features. Unfortunately, no single molecular marker exists to reliably differentiate melanoma from benign lesions such as nevi. This study explored the potential of autofluorescent molecules within tissues to provide molecular fingerprints indicative of degenerated melanocytes in melanoma.

Methods: Using hyperspectral imaging (HSI) and spectral phasor analysis, we investigated autofluorescence patterns in melanoma compared to intradermal nevi. Using UV excitation and a commercial spectral confocal microscope, we acquired label-free HSI data from the whole-slice samples.

Results: Our findings revealed distinct spectral phasor distributions between melanoma and intradermal nevi, with melanoma displaying a broader phasor phase distribution, signifying a more heterogeneous autofluorescence pattern. Notably, longer wavelengths associated with larger phases correlated with regions identified as melanoma by expert dermatopathologists using H&E staining. Quantitative analysis of phase and modulation histograms within the phasor clusters of five melanomas (with Breslow thicknesses ranging from 0.5 mm to 6 mm) and five intradermal nevi consistently highlighted differences between the two groups. We further demonstrated the potential for the discrimination of several melanocytic lesions using center-of-mass comparisons of phase and modulation variables. Remarkably, modulation versus phase center of mass comparisons revealed strong statistical significance among the groups. Additionally, we identified the molecular endogenous markers responsible for tissue autofluorescence, including collagen, elastin, NADH, FAD, and melanin. In melanoma, autofluorescence is characterized by a higher phase contribution, indicating an increase in FAD and melanin in melanocyte nests. In contrast, NADH, elastin, and collagen dominate the autofluorescence of the nevus.

Discussion: This work underscores the potential of autofluorescence and HSI-phasor analysis as valuable tools for quantifying tissue molecular fingerprints, thereby supporting more effective and quantitative melanoma diagnosis.

Keywords: cancer; fluorescence microscopy; hyperspectral imaging; melanoma; nevus; phasor analysis; skin cancer; spectral phasor.

Figure 1

Phasor analysis pipeline. Images of increasing lambda steps from the HSI stack. If the image dimension is $\text{d}=\text{m}\times \text{n}$ , where $\text{d}$ is a positive integer, then there are $\text{d}$ spectra in the HSI stack. After we apply the phasor transform (represented by $\text{T }$ ), there are $\text{d}$ pairs of coordinates $(\text{G},\text{ S})$ in the phasor plot. In (A), it is possible to see that a single pixel in the lambda stack has an associated spectrum, which is transformed and represented in the phasor. The phasor plot allows us to identify how fluorescent components are distributed over the phasor space. Similar spectra have $(\text{G},\text{ S})$ coordinates close to each other, whereas different spectra are separated. Using the reciprocity principle, it is possible to select regions of interest (ROI) in the phasor plot, such as the blue circle, and color the related pixels of the ROI over a gray image. Selecting an ROI in the gray image also enables us to obtain phasor coordinates for this ROI. Each $(\text{G},\text{ S})$ pair has an associated pair of phase and modulation (ϴ, ϱ); based on that, we can create a color scale and obtain a pseudocolor image. An example is shown in the figure on the right side, where we implemented a color scale where low phases correspond to green and high phases correspond to blue. (B) Simulation of spectral phasor properties (changes in the spectrum center of mass, phase shift, spectrum width (modulation), and linear combination).

Figure 2

Pipeline for the evaluation of skin melanocytic lesions. (A) A minimal margin was used to excise the pigmented lesion, and then the sample was split using the “loaf bread” technique for the traditional and new HSI protocols. The black dashed lines represent the line of lesion excision, and the red dashed lines are the “loaf bread” cut. (B) The current standardized diagnosis of melanoma is based on histological studies aided by immunohistochemistry. (C) We propose a novel approach based on label-free slices and HSI images using spectral phasor analysis. H&E, hematoxylin and eosin; IHQ, Immunohistochemistry; HSI, Hyperspectral imaging.

Figure 3

Comparison of nevus and melanoma anatomo-pathology using H&E and label-free HIS-phasor plots. (A, E) Hematoxylin–eosin images of the nevus and invasive melanoma samples. (B, F) Label-free average intensity of HSI obtained from the previous two tissue blocks (A, E, respectively). (C, G) Phasor plots obtained from the HSI images in (B, F) respectively. The inserted color scale represents the color scheme used to generate a pseudocolor image based on the phase of each pixel. (D, H) Pseudocolor images obtained by using the color scale shown in (C, G). Scale bars represent $500\text{ μm}$ .

Figure 4

Analysis of phasor distribution in the region of interest associated with injuries. (A, B) Regions segmented where melanocytic injuries were present. An expert dermatologist identified the regions of interest. (C, D) Phasor plots corresponding to these regions. (E) shows the modulation distribution for the nevus and melanoma regions, whereas (F) represents the distribution of the phase.

Figure 5

Study of the modulation and phase distribution obtained from the patients in Table 1 with intradermal nevi or invasive melanomas. (A, B) Histograms of modulation and phase, respectively, for nevus and melanomas. (C, D) present the center of mass of each distribution. Those groups are statistically described with the average and standard deviation as follows: the nevi group has x = 0.47 ± 0.03 for the modulation and x = 108 ± 6 for the phase. In comparison, the melanoma group has x = 0.58 ± 0.02 for the modulation and x = 88 ± 3 for the phase. A p = 0.0003 was obtained for the modulation group, while p = 0.0008 was obtained for the phase group. (E) presents two groups of data obtained when plotting the modulation vs. phase. It also has the confidence ellipses obtained for each group, built with a standard deviation of σ = 68.5%, 2σ = 95.5%, and 3σ = 99.7% each one.

Figure 6

Spectral phasor fingerprints for autofluorescent biomolecules in the skin. (A) Pure component centroid obtained with the k-means algorithm using HSI data acquired experimentally from independent samples listed in (B). (B) Average emission spectrum for collagen I, II, III, IV, elastin, NADH, FAD, and melanin.

Figure 7

Nests of melanocytes were imaged at ×63 from a nevus and an invasive melanoma. It shows the phasor analysis for these regions and the distribution of the previously obtained pure components. (A, D) Images show the label-free average intensity image of the HSI stack; it can appreciate nests of benign and atypical melanocytes, respectively. Images (B, E) show the corresponding phasor plot and the centroid of each molecular fingerprint overlaid. Scale bars are $50\text{µm}$ . (C, F) present pseudocolor images of the nevus and melanoma, respectively. (G) shows the modulation histogram, and (H) shows the phase histogram for both samples.