Oblique-incidence spatially resolved diffuse reflectance spectroscopic diagnosis of skin cancer

This paper presents the use of spatially resolved oblique-incidence diffuse reflectance spectroscopy for skin cancer diagnosis. Spatio-spectral data from 166 pigmented skin lesions were collected for the wavelength range from 455 to 765 nm. A set of neural network based classifiers separates the pigmented malignant melanomas from the benign and dysplastic subgroups. A total of 110 lesions were used as the training set and 56 lesions were used as the testing set. This classifier performs with an overall 100% sensitivity and 92% specificity for the training set and 100% sensitivity and 88% specificity for the testing set. The second classifier was designed to separate the benign from the dysplastic subgroups. For the second classifier a total of 100 lesions were used as the training set and 51 lesions were used as the testing set. The overall classification rates were 94% and 88% for the training and testing sets respectively.