NONINVASIVE DERMATOLOGICAL LESION CLASSIFIER

皮肤癌是当今美国增长最快的癌症。

Skin cancer is the fastest growing cancer in the United States today. Approximately 34,100 Americans developed cutaneous melanoma in 1995, and 7,200 died of the disease; of the survivors, many must contend with the ongoing trauma of disfigurement and fear. Skin biopsies are now the most frequently performed medical procedure reimbursed by Medicare. It is axiomatic among dermatologists that early detection and diagnosis are critical in the care and treatment of skin cancer patients. Great strides have been made in recent years in early detection of suspect skin lesions; however, the diagnosis remains based in the subjective evaluation of which skin lesions to biopsy. This decision is the basis of a great dilemma for physicians of at-risk patients who develop literally hundreds of lesions which could be pre-cancerous or cancerous. On one hand biopsies are expensive and traumatic; on the other, failure to biopsy the right lesion can lead to severe consequences. The dilemma is further exacerbated by the fact that 50-80% of biopsies prove unnecessary after the fact, contributing to an enormous of valuable health care dollars, patient trauma and negative patient behavior feedback. Recent developments in dermatological spectroscopy used to train an artificial neural net technology suggest that an automated clinical diagnostic aid which produces a quantitative rather than qualitative diagnostic assessment of skin lesions is possible. This project proposes development and testing of such a product. Spectroscopic samples of approximately 500 patients with abnormal skin lesions will be coupled with an equal number of normal skin spectra and used to train an artificial neural net classifier. This automated diagnostic aid will be tested against a large number of test samples for which a histological diagnosis is available for evaluation of the system. PROPOSED COMMERCIAL APPLICATIONS: The proposed project will lead to a non-invasive, in-office, real-time test to provide an automated, repeatable diagnostic probability of the nature of skin lesions prior to biopsy. Skin biopsies are now the most frequently performed reimbursed Medicare procedure, and as many as 50-80% are found not to be necessary after the fact. The low cost of this test, and rapid amortization of the system, coupled with the enormous health care cost savings possible in conjunction with a significant and widely recognized health problem, suggest that this product could have great commercial potential.