Quantitative Analysis of Skin Moles
The Skin Of Mine Mole Analysis module provides quantitative feedback about symmetry, border, and color regularity of moles from ordinary digital photos. The graph shown below is the analysis for the mole shown in the photo. This mole was later biopsied and determined to be benign.
Interesting, but how well does the analysis work? One way to answer that question is to define a test set containing an assortment of moles and assign a “truth” value to each test mole.
We can then compare the automatic analysis results to the “truth” for each case, and amass statistics over items in the test set.
The truth value could be a hand-labeled value of symmetry, border, and color regularity, assigned by an expert such as a dermatologist.
Another method for establishing “truth” would be to look at biopsy results for each of the moles. Although benign moles are not always symmetric with regular borders and colors, and melanoma is not always irregular in one or more of those dimensions, there is a tendency in those directions.
Because we know that melanoma is correlated with irregularities, we can examine the correlation between the set of automatically generated labels and the set of biopsy results. We’ve taken a step in that direction with the following: please click on MoleAnalysis2 to read the memo.