Diagnosis of Melanoma Based on Data Mining and ABCD Formulars

A parameter called TDS (Total Dermatoscopic Score), calculated by the wellknown ABCD formula, is frequently used in melanoma diagnosis. In our previous research we found a new formula, similar to the original ABCD formula, that yielded fewer diagnostic errors. This new ABCD formula was developed using data mining techniques, in particular, the rule induction algorithm LEM2, a part of the data mining system LERS. In this paper we compare the quality of the old and new ABCD formulas, measured by the number of diagnostic errors, using three other data mining techniques: two rule induction algorithms, LEM1 and a modified version of LEM2 called MLEM2, and the decision tree generating system C4.5. Additionally, we compare the quality of diagnosis using TDS (original and new) and diagnosis without using TDS at all, to address complaints by some diagnosticians that TDS does not improve diagnosis of melanoma. Our experiments show that TDS is a valuable tool, significantly increasing melanoma diagnosis accuracy.