Optical Spectroscopy Shows Limited Evidence of Predicting Skin Cancer
A pilot study suggests that the noninvasive technique may not be ready to replace biopsies yet.
Biopsies are routinely used to screen skin lesions suspected to be cancerous, but the invasive procedure can cause bleeding, infection, unsightly scars and other complications. One potential noninvasive alternative is optical spectroscopy, which determines the scattering and absorption properties of tissue. The method can also sample multiple sites without the need for tissue removal, provide nearly immediate feedback, and identify subtle biophysical features that precede more obvious morphological changes.
In a pilot study published Jan. 13 in the journal Lasers in Surgery and Medicine, researchers used optical spectroscopy to characterize the optical properties of 27 skin lesions collected from 18 adult patients. These lesions included scar tissue, moles, precancerous patches of scaly skin, and cancerous tissue. The spectroscopy markers of interest included scattering, hemoglobin saturation, total hemoglobin concentration, and eumelanin concentration.
For each lesion, one of two physicians used the optical measurements to predict the probability of malignancy on a scale from 0 to 100 percent, prior to biopsies. The researchers then compared the predictions to biopsy results to evaluate the clinical utility of optical spectroscopy in place of biopsies as a tool for screening skin cancer.
On average, an absolute difference of 29 percent was observed between pre‐biopsy predictions and biopsy results. While 12 of the lesions were correctly predicted to within 5 percent certainty of benign or malignant classification, five lesions demonstrated a difference of 75 percent or greater. Overall, clinical predictions demonstrated an estimated sensitivity of 92 percent and a specificity of 54 percent.
Although the study was not statistically conclusive, their findings justify the practice of obtaining biopsies even when clinical suspicion of malignancy is low, at least for the time being. In the future, machine learning algorithms that combine multiple spectroscopic parameters may be able to help distinguish benign from malignant tissue types and help clinicians decide whether or not to proceed with biopsy or continue observation.