640 A hyperspectral index-based approach for in vivo automatic detection of skin tumors from hyperspectral images

Abstract

Skin cancer is the most common human malignancy and early diagnosis is important for
successful treatment. Hyperspectral imaging provides both spectral and spatial information which
may be useful for tumor diagnosis. In the present study, hyperspectral images of 36 skin tumors
(seborrheic keratosis, actinic keratosis, basal cell carcinoma, squamous cell carcinoma) were
acquired and analyzed. A skin cancer index was defined taking into account the differences in the
angles made by the spectral slopes of the tumors and normal skin and the x axis in the spectral range
(580-600) nm (where the differences in spectral signatures proved to be significant). When compared
to histopathological diagnosis, the index allowed for differentiation between benign and malignant
tumors and even between seborrheic and actinic keratosis or basal cell carcinoma and squamous
cell carcinoma. Benign tumors had an index under 0.39999, while basal cell carcinoma ranged
between 0.4000-0.59999, and squamous cell carcinoma between 0.6000-06999. In conclusion, the
hyperspectral image-based skin tumor index is an objective method of identification and
differentiation of skin tumors and may become a useful tool in their early diagnosis
Keywords: skin cancer hyperspectral index, reflectance, slope, seborrheic keratosis, actinic keratosis, basal cell carcinoma, squamous cell carcinoma