B. Farina et al., Multispectral imaging approach in the diagnosis of cutaneous melanoma: potentiality and limits, PHYS MED BI, 45(5), 2000, pp. 1243-1254
In an attempt to overcome the subjectiveness of clinical observation in the
diagnosis of cutaneous melanoma, a computerized method is proposed. Reflec
tance images of 237 pigmented lesions (67 melanomas and 170 non-melanomas)
were analysed using a telespecificphotometric technique. This device consis
ts of a CCD camera with 17 interference filters. Images were acquired at se
lected wavelengths, from 420 to 1040 nm. Morphological and reflectance rela
ted parameters were extracted from the wavelength-dependent images of the l
esions. The most significant features in the comparison between benign and
malignant lesions were: lesion dimension (P < 10(-8) at 578 nm); mean value
(P < 10(-7) at 940 nm) and standard deviation(P < 10(-4) at 904 nm) of les
ion reflectance; lesion roundness (P < 10(-5) at 461 nm); and border irregu
larity (P < 10-4 at 461 nm). Based on these parameters, a discriminant func
tion between the two populations of lesions (naevi and melanomas) was obtai
ned. By using the results of the analysis of the recruited lesions as 'trai
ning data', discriminant functions enabled the assignment of a score, or a
'risk probability', to each studied lesion. By imposing a sensitivity of 80
% (a figure that mimics the diagnostic capability of an experienced clinici
an), entering or not entering the lesion dimension as input data in the dis
criminant analysis led to a specificity of 51% or 46% respectively.
The high number of false-positive cases, which is a consequence of the sele
ction criteria of the lesions, is, at present, the major limitation of the
current technique. Nevertheless, our results suggest that an imaging-based
computer-assisted device could be capable of discriminating malignant lesio
ns mainly by evaluation of reflectance, especially in the infrared region,
and shape properties. The dimension of a lesion should not be essential in
the diagnosis of melanoma and, in our opinion, small melanomas should be re
cognized by a computer system as well as they are on clinical grounds.