Introduction

Malignant melanoma (MM) is a lethal but curable skin cancer. It has shown one of the highest rates of increase in the last 20 years. The incidence and mortality continue to rise in many countries. (La Vecchia et al., 1999; Jemal et al., 2001; De Vries et al., 2003; Pearce et al., 2003). Although the seven-point checklist and ABCDE rule are successful in alerting the public to potential melanoma, their use to physicians in differentiating melanoma from other pigmented lesions is limited (Higgins et al., 1992; Mallet et al., 1993; Healsmith et al., 1994).

Early detection is the basis for reducing the mortality rate (Breslow, 1970) and non-invasive skin imaging plays an expanding role in this regard. Digital dermoscopy and spectral imaging devices (SIAscopy) have been experimentally shown to improve melanoma sensitivity and help reduce unnecessary excision of benign nevi (Moncrieff et al., 2002; Rubegni et al., 2002). Optical methods, however, are less promising for other common benign-pigmented lesions (BPLs) such as seborrheic keratoses (SKs) (Marchesini et al., 1991; Wallace et al., 2000). This may hamper the extension of automated optical devices to primary care, where, in practice, a much larger range of benign pigmented lesions is seen. Three-dimensional high-resolution ultrasound (HRU) could be a cheap and useful adjunct to existing optical methods in achieving this goal. HRU has been shown to differentiate between melanoma and SKs, and less so between melanoma and benign nevi (Harland et al., 2000).

Skin ultrasonography was first described by Alexander and Miller in 1979. A transducer directs high-frequency (>15 MHz) wave pulses into the skin and detects returned echoes at an appropriate time delay. The backscattered signal is the basis of all measurements made. The strength of this signal varies with strength of incident signal, depth beneath the skin's surface, scatter properties, and absorption characteristics of the lesion. The echoes therefore carry information on the absorption and reflective properties of tissue. Image generation by this method has been described previously (Kreitz, 1992; Rallan and Harland, 2003).

Skin HRU has proven to be an effective method for non-invasive measurements and disease monitoring, but few authors have reported quantitative analysis in relation to tissue characterization (Edwards et al., 1989). In a study of 70 tumors, differences in acoustic characteristics have been quantified for MM, benign nevi, and SKs using two-dimensional B-mode HRU. Significant differences were particularly noted in the attenuating characteristics between pigmented lesions (Harland et al., 2000). These findings suggest that the biological nature of a tumor may affect the extent to which high-frequency sound waves are transmitted through it. Melanoma was reported to be less attenuating than SKs and, to a lesser extent, nevi.

Enhancing the measurement of attenuation, by reducing errors, especially the variance, could improve quantitative discrimination among pigmented lesions and potentially reduce the referral of benign lesions from the community. To achieve this objective, a three-dimensional HRU technique termed reflex transmission imaging (RTI) was used. Although originally described for internal organ imaging (Green and Arditi, 1985), the method has been refined (Rallan et al., 2006) and is employed here for in vivo skin imaging. A reflex transmission image is generated using a strongly focused transducer (whose focus lies at the skin surface) and integrating volume data from the retro-lesional dermis (Figure 1). This produces an image for which the signal is predominantly influenced by ultrasonic attenuation in the focal plane. Similarly, integration zones can be selected through the lesion body and the through the skin surface. The former was used to generate a "lesional backscatter image" (LBI) and the latter an "entry echo image" (EEI). Thus, while RTI parameters refer to lesion attenuation properties, LBI and EEI parameters depict intra-lesional sound reflection and surface sound reflectance characteristics, respectively.

Figure 1.

Ultrasound beam and image planes for three image types. The RTI is an attenuation image, LBI is a lesion reflection image, and EEI is a surface reflectance image. The transducer moves in two horizontal directions and data are integrated from multiple horizontal slices. The images therefore depict three-dimensional volume data.

Full figure and legend (104K)

The aim of this study was to test the hypothesis that attenuation measured in this way could produce differentiation of common BPLs from melanoma and to assess if other derivable parameters (from LBI and EEI) could add to this discrimination.

Results

Twenty-five MM (18 female, seven male) and 62 non-cancerous lesions (45 female, 17 male) in patients aged 21–67 years were analyzed. Eleven melanomas were in situ and 14 invasive (mean Breslow thickness 0.970.29 mm, range 0.25–2.0 mm). Non-cancerous lesions consisted of SKs (n=24) and benign pigmented lesions (BPL, n=38). The most frequently encountered tumors in the BPL category were melanocytic nevi (20 compound nevi, nine junctional nevi, and three intradermal nevi). Five benign lentigos and one dermatofibroma were also included.

Statistics

The aims of the statistical analysis were, firstly, to compare the means of all six features among the three groups to identify those features that may aid discrimination (inter-group analysis), and to identify if two quantitative parameters significantly correlated within a group (intra-group analysis). The latter may help elucidation of sound behavior in different tumors. Correlations were sought among the six relative measures by calculating Pearson's coefficient (r).

Each lesion category was treated as an independent sample. The relevant discrimination sought was between MM and SK (pair 1) and between MM and BPL (pair 2). The differences in means of parameters between samples were compared using unpaired Student's t-test. Statistical tables and calculations were generated in SPSS v10.

Comparison of melanoma and SK

Table 1 lists the mean and standard deviation (SD) of all six parameters for MM and SKs. P-values are listed in the last row. SK showed a significant quantitative difference from melanoma on three features, exhibiting greater surface heterogeneity (EEI; P=0.002), and melanoma a greater intra-lesional heterogeneity (LBI; P=0.006). Finally, the difference in RTI contrast demonstrates that SKs are considerably more attenuating than melanomas (P<0.001).

Table 1 - Mean (SD) of all six parameters for malignant melanoma and SK.

Full table

Comparison of melanoma and other BPLs

Table 2 lists the values of the six parameters with P-values for the difference in means between melanoma and other BPLs. Melanoma are more attenuating than other BPLs (predominantly benign nevi) and also have greater surface heterogeneity, but a lower intra-lesional heterogeneity (EEI and LBI relative heterogeneity, respectively). These key differences among melanoma, SKs and other BPLs are summarized in Figure 2. Examples of the image appearances of these lesion types are shown in Figure 3.

Figure 2.

Quantitative differences between melanoma, benign melanocytic SKs, and other benign-pigmented lesions (BN) on three parameters. The dashed lines mark the value ranges beyond which melanoma was excluded. The error bars show the 95% confidence interval about the mean. Exclusion was one tailed in order to assess the most conservative result for specificity. The category labelled as BN above contained five benign lentigos and one dermatofibroma as well.

Full figure and legend (136K)

Figure 3.

Examples of the appearance of the three lesion types and the three scan types are shown. Although SKs can sometimes be recognized on RTI by visual inspection, we noted this is not the case with acanthotic keratoses. Similarly, lesions in the BPL category cannot be reliably differentiated from melanoma upon visual inspection of the images. Although elusive to the human eye, the acoustic differences were numerically quantifiable making differentiation entirely objective.

Full figure and legend (138K)

Table 2 - Mean (SD) of all six parameters for malignant melanoma and other benign pigmented lesions.

Full table

Quantitative discrimination

The usefulness of a potentially differentiating parameter depends upon the degree of separation produced. With reference to Figure 2, a sensitivity of 100% can be set; a quantitative value beyond this threshold excluded all melanoma seen in this study. Thus in comparing SKs and melanoma, RTI contrast and LBI relative heterogeneity both have a specificity of 38% (nine out of 24 SKs were classified correctly by each parameter) when sensitivity is set to 100%. Similarly, EEI relative heterogeneity classifies 29% (seven out of 24) SKs correctly without missing melanoma. Furthermore, each parameter excludes different cases so that the maximum specificity with all three combined is about 79%.

In the comparison of melanoma and other BPLs, EEI relative heterogeneity had the highest specificity of 30% followed by LBI relative heterogeneity (15%) and RTI contrast (10%). The overall specificity of the three parameters for classifying nevi correctly is around 55% again with sensitivity for melanoma at 100%.

From a total of 62 non-cancerous lesions, 40 (19 in the SK group and 21 in the BPL group) were correctly identified as benign. Thus, potentially, 65% of patients referred with benign lesions could be safely reassured and referral avoided.

With reference to Figure 4, no within-group correlation was identified among the three parameters used above. However, surface reflectance correlates with surface heterogeneity for SKs (P<0.001) and intra-lesional contrast correlates with intra-lesional heterogeneity for all groups (P<0.05).

Figure 4.

Inter- and intra-group comparison of acoustic parameters. MM – malignant melanoma, SK- seborrheic keratoses, BN – other benign-pigmented lesions. (a) Significant correlation between surface reflectance and surface heterogeneity exists for the SK group but not for the BN and MM groups. (b) Intra-lesional contrast correlates with intra-lesional heterogeneity in all three groups. The relationship is weakest for MM. (c) Heterogeneity of attenuation increases with overall tumor attenuation. The relationship is weakest for the BN group.

Full figure and legend (207K)

Discussion

This study has confirmed that significant acoustic differences exist between common BPLs and MM. The high attenuation and prominent entry echo of some SKs has been reported previously (Harland et al., 2000). This can be explained by the surface keratinization of these tumors, which makes them reflective. Keratin deposition also renders the surface irregular, so that the surface contrast was found to correlate with surface heterogeneity in this group only (Figure 4a). Relative surface reflectance of SKs (EEI contrast) was higher than that of melanoma, but this difference was not significant (P=0.229). Similarly, intra-lesional reflection (LBI contrast) does not differ greatly. Sound absorption, rather than reflection, may explain the difference in attenuation. The increased intra-lesional heterogeneity of SKs versus melanoma is a new finding, and appears to be discriminatory.

The use of multiple features in this study produced a specificity of 79% and sensitivity of 100% for the differentiation of SKs from melanoma. This specificity is similar to that achieved in a previous report (Harland et al., 2000) despite the fact that previously acanthotic SKs were excluded, whereas in this study, they were included.

Between melanoma and other BPLs/benign nevi, surface heterogeneity was the strongest discriminator with melanoma showing greater surface heterogeneity. This parameter alone produced a specificity of 30% (sensitivity 100%). These trends are similar to those shown previously by Harland et al., 2000. Inclusion of EEI relative heterogeneity and RTI contrast however improved specificity to 55%.

Discrimination is stronger between SKs and melanoma than between melanoma and other BPLs (predominantly benign nevi). Only those parameters that showed a significant difference in means between groups were employed. Thus out of six features, three were treated as potential differentiators. No single parameter produces a high specificity, although this improves when features are combined. No conclusion, however, can be drawn on their use regarding a wider range of pigmented lesions representative of general practice.

Whereas optical methods provide objective data on the surface features of pigmented lesions, HRU offers quantitative analysis of sub-surface and sub-lesional characteristics. Thus a combined approach may, by introducing further independent variables, improve classification. A potential 65% reduction in referral of benign lesions is projected from this sample. The majority of lesions excluded were SKs, but more than half the other BPLs studied were also differentiated from a sample of relatively thin melanomas. Thus, in conjunction with optical aids, early detection and considerable streamlining of referrals from the community may be achievable. However, detection criteria were arbitrarily set to achieve 100% sensitivity for this data set and a prospective application of these criteria may or may not produce the same specificities and reduction in referral rate quoted above.

Quantitative methods may allow differentiation between pigmented tumors by an entirely objective method and without the need for specialized skill and training in image interpretation. Future devices with appropriate software can be made compact (lap-top size). Low-cost complimentary objective methods in the community may be a step towards curbing the mortality caused by melanoma particularly with the possibility of a relentless rise in incidence in many countries (Giles et al., 1996; Diffy, 2004). Although the accuracy of individual automated digital devices is becoming established for equivocal melanocytic lesions in dermatology clinics, further studies need to be conducted in the community, where a larger range of pigmented lesions such as SKs pose diagnostic difficulties.

Materials and Methods

Ultrasound equipment

A single-element commercial skin scanner (Dermascan Cv3 Cortex ApS, Denmark), which scans in two dimensions (x–y), was used. The scanner was modified for RTI by using a transducer with an f-number of 0.95. A depth penetration of 7 mm was considered appropriate in order to acquire data from a sufficient volume of skin on most body sites. Cellular detail was not required and 20 MHz (resolution 60–120 m) was therefore used as the sound wave frequency. The field of view of the modified scanner was 22.4 22.4 mm. The average speed of sound in human skin was taken as 1,600 m/s. A total of 224 cross-sections (B-frames) were acquired in the lateral direction with a step increment of 0.1 mm; each B-frame was made of 224 lines also with a step increment of 0.1 mm. Further details of the data-acquisition and image construction methods have been presented previously (Rallan et al., 2006).

Patient recruitment

The study was approved by the local research ethics committees. Patients were recruited via hospital-based skin cancer screening clinics in the South London region to which they are referred with suspected skin cancer. Only patients referred with a suspicion of melanoma were considered. All patients subsequently given a clinical diagnosis of SK, benign nevus, or with a clinical suspicion of melanoma by a consultant dermatologist were eligible. The following exclusion criteria were applied:

1. Tumors on the head and neck – excluded owing to size of the scanner head of the prototype ultrasound equipment.
2. Lesions larger than 2 cm in maximum diameter – excluded owing to limit of scanning window.
3. Cosmetic considerations where surgical removal (of clinically benign lesion) was considered inappropriate.

Verbal and printed information was provided and written consent obtained with adherence to the Declaration of Helsinki Principles.

The lesion was removed under local anesthetic following data acquisition. Histological diagnosis was then used to classify the lesion in one of three groups, MM, SK, or other BPL. In cases of histological atypia or dyplasia, suggesting but not confirming melanoma, the lesion was classed in accordance with the clinical management protocol (usually as melanoma).

Derivation of quantitative parameters

For every lesion, three images (RTI, LBI, and EEI) were generated. Lesion boundaries are not reliably discernible on the ultrasound images. Optical boundaries from corresponding digital photographs were therefore used. The optical boundary was manually drawn, and following registration of optical with ultrasound images, the boundary was transferred on to the ultrasound images. Details of image registration and processing are described elsewhere (M Dickson, PhD Thesis, University of London, 2004). Two quantitative features were calculated from each image (Figure 5).

Figure 5.

RTI and corresponding photograph of a melanoma are shown. E_o – mean echo value in normal skin, E_i – mean echo value within tumor boundary, H_o – SD of echoes in normal skin, and H_A – SD of values within tumor boundary. The lesion is not visible on this RTI and is defined using the optical boundary of the transformed photograph (right). Lesion contrast is calculated as (E_o–E_i)/E_o. Similarly, relative heterogeneity is calculated as (H_o–H_A)/H_o. Marker and bubble artifacts are excluded from calculations. The two features were calculated on EEI and LBI as well for each lesion.

Full figure and legend (58K)

Lesion contrast

This is defined as a relative measure of mean echo strength within the lesion boundary compared to that of surrounding normal tissue (used as a control). With reference to Figure 2, lesion contrast (C_L) is calculated as

where E_i refers to mean echo value within the lesion area and E_o is the mean outside the lesion boundary.

Relative lesion heterogeneity

Heterogeneity is defined as the non-uniformity of echo strength within an outlined region (a selected group of pixels). The SD of echo values within a region is taken as a quantitative measure of heterogeneity. The greater the SD, the greater the absolute sonographic heterogeneity (H_A) which is calculated as

where SD_L is the SD of echoes within the drawn lesion boundary, E_p is the echo value from each pixel and n is the number of pixels. Relative heterogeneity (H_R) compares lesional heterogeneity to that of surrounding normal skin used as a control

where H_o refers to the background SD outside the region of interest.

Conflict of Interest

The authors state no conflict of interest.

References

1. Alexander H, Miller DL (1979) Determining skin thickness with pulsed ultrasound. J Invest Dermatol 72:17–19 | Article | PubMed | ChemPort |
2. Breslow A (1970) Thickness, cross-sectional area and depth invasion in the prognosis of cutaneous melanoma. Ann Surg 172:902–908 | PubMed | ISI | ChemPort |
3. De Vries E, Bray FI, Coebergh JW, Parkin DM (2003) Changing epidemiology of malignant cutaneous melanoma in Europe 1953–1997: rising trends in incidence and mortality but recent stabilisations in Western Europe and decreases in Scandinavia. Int J Cancer 107:119–126 | Article | PubMed | ISI | ChemPort |
4. Diffy BL (2004) The future incidence of cutaneous melanoma within the UK. Br J Dermatol 151:868–872 | PubMed |
5. Edwards C, Al-Boosi MM, Marks R (1989) The use of A-scan ultrasound in the assessment of small skin tumours. Br J Dermatol 121:297–304 | Article | PubMed | ChemPort |
6. Giles GG, Armstrong BK, Burton RC, Staples MP, Thursfield VJ (1996) Has mortality from melanoma stopped rising in Australia? Analysis of trends between 1931 and 1994. BMJ 312:1121–1125 | PubMed | ChemPort |
7. Green PS, Arditi M (1985) Ultrasonic reflex transmission imaging. Ultrasonic Imag 7:201–214 | ChemPort |
8. Harland CC, Kale SG, Jackson P, Mortimer PS, Bamber JC (2000) Differentiation of common benign pigmented lesions from melanoma by high resolution ultrasound. Br J Dermatol 143:281–289 | Article | PubMed | ISI | ChemPort |
9. Healsmith MF, Bourke JF, Osborne JE, Graham-Brown RAC (1994) An evaluation of the revised seven-point checklist for the early diagnosis of cutaneous malignant melanoma. Br J Dermatol 130:48–50 | Article | PubMed | ChemPort |
10. Higgins EM, Hall P, Todd P, Murthi R, Du Vivier AW (1992) The application of the seven point checklist in the assessment of benign pigmented lesions. Clin Exp Dermatol 17:313–315 | Article | PubMed | ChemPort |
11. Jemal A, Devesa SS, Hartge P, Tucker MA (2001) Recent trends in cutaneous melanoma incidence among whites in the United States. J Natl Cancer Inst 93:678–683 | Article | PubMed | ChemPort |
12. Kreitz G (1992) A- and B-scan techniques of conventional ultrasound units. In: Ultrasound in Dermatology. (Altmeyer P, El Gammal S, Hoffman K, eds). Berlin: Springer Verlag, 12–21
13. La Vecchia C, Lucchini F, Negri E, Levi F (1999) Recent declines in worldwide mortality from cutaneous melanoma in youth and middle age. Int J Cancer 81:62–66 | Article | PubMed | ChemPort |
14. Mallet RB, Fallowfield ME, Cook MG, Landells WN, Holden CA, Marsden RA (1993) Are pigmented lesion clinics worthwhile? Br J Dermatol 129:689–693 | Article | PubMed |
15. Marchesini R, Brambilia M, Clemente C, Maniezzo M, Sichirollo AE, Testori A et al. (1991) In vivo spectrophotometric evaluation of neoplastic and non-neoplastic skin pigmented lesions-1. Reflectance measurements. Photochem Photobiol 53:77–84 | PubMed | ISI | ChemPort |
16. Moncrieff M, Cotton S, Claridge E, Hall P (2002) Spectrophotometric intracutaneous analysis: a new technique for imaging pigmented skin lesions. Br J Dermatol 146:448–457 | Article | PubMed | ISI | ChemPort |
17. Pearce MS, Parker L, Cotterill SJ, Gordon PM, Craft AW (2003) Skin cancer in children and young adults: 28 years' experience from the Northern Region Young Person's Malignant Disease Registry, UK. Melanoma Res 13:421–426 | Article | PubMed |
18. Rallan D, Dickson M, Bush NL, Harland CC, Mortimer P, Bamber JC (2006) High-resolution ultrasound reflex transmission imaging and digital photography: potential tools for the quantitative assessment of pigmented lesions. Skin Res Technol 12:50–59 | Article | PubMed | ChemPort |
19. Rallan D, Harland CC (2003) Ultrasound in dermatology – basic principles and applications. Clin Exp Dermatol 28:632–638 | Article | PubMed | ChemPort |
20. Rubegni P, Burroni M, Cevenini G, Perotti R, Dell'Eva G, Barbini P et al. (2002) Digital dermoscopy analysis and artificial neural network for the differentiation of clinically atypical pigmented skin lesions: a retrospective study. J Invest Dermatol 119:471–474 | Article | PubMed | ISI | ChemPort |
21. Wallace VP, Crawford DC, Mortimer PS, Ott RJ, Bamber JC (2000) Spectrophotometric assessment of pigmented skin lesions: methods and feature selection for evaluation of diagnostic performance. Phys Med Biol 45:735–751 | Article | PubMed | ChemPort |