The rising incidence of skin cancer has led to an increase in the number of patients with skin lesions that require diagnosis, mostly using subjective visual examination. Successful treatment depends on early diagnosis. Unfortunately diagnostic accuracy, even by experts, can be as low as 56%; therefore, an accurate, objective diagnostic aid is greatly needed. Reflectance characteristics of pigmented skin lesions were documented to evaluate their diagnostic potential. Reflectance spectra in the wavelength range 320-1100 nm were obtained from 260 lesions. Differences between spectra from benign and malignant lesions were utilized by extracting features with the best discriminating power. Discrimination was evaluated using two techniques: multivariate statistical analysis and artificial neural networks, using histology as the standard. Each technique was tested in a blind study and assessed in terms of its ability to diagnose new cases and compared to the clinical diagnosis. The artificial neural network achieved the best diagnostic performance for discriminating between malignant melanoma and benign nevi, having a sensitivity of 100% and a specificity of 65%. Utilization of visible and infrared techniques for monitoring skin lesions has lead to improvements in diagnostic accuracy. We conclude that these techniques are worthy of further development and evaluation in clinical practice as a screening tool.
|Number of pages||9|
|Journal||Proceedings of SPIE - The International Society for Optical Engineering|
|Publication status||Published - 2002|