A multiple classifier system for early melanoma diagnosis

• Published in: Artificial intelligence in medicine Vol. 27; no. 1; pp. 29 - 44
• Main Authors: Sboner, Andrea, Eccher, Claudio, Blanzieri, Enrico, Bauer, Paolo, Cristofolini, Mario, Zumiani, Giuseppe, Forti, Stefano
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 2003
• Subjects: Algorithms; Classifier combinations; Cost-sensitive learning; Databases as Topic; Decision Making, Computer-Assisted; Decision support system; Decision Support Systems, Clinical; Decision Trees; Discriminant Analysis; Early melanoma diagnosis; Humans; Image Processing, Computer-Assisted; Machine learning; Mathematical Computing; Melanoma - classification; Melanoma - diagnosis; Models, Statistical; Sensitivity and Specificity; User-Computer Interface; Cost-sensitive learning; Early melanoma diagnosis; Decision support system; Machine learning; Classifier combinations
• ISSN: 0933-3657; 1873-2860
• DOI: 10.1016/S0933-3657(02)00087-8

Melanoma is the most dangerous skin cancer and early diagnosis is the key factor in its successful treatment. Well-trained dermatologists reach a diagnosis via visual inspection, and reach sensitivity and specificity levels of about 80%. Several computerised diagnostic systems were reported in the literature using different classification algorithms. In this paper, we will illustrate a novel approach by which a suitable combination of different classifiers is used in order to improve the diagnostic performances of single classifiers. We used three different kinds of classifiers, namely linear discriminant analysis (LDA), k-nearest neighbour ( k-NN) and a decision tree, the inputs of which are 38 geometric and colorimetric features automatically extracted from digital images of skin lesions. Multiple classifiers were generated by combining the diagnostic outputs of single classifiers with appropriate voting schemata. This approach was evaluated on a set of 152 digital skin images. We compared the performances of multiple classifiers (2- and 3-classifier groups) between them and with respect to single ones (1-classifier group). We further compared the classifiers’ performances with those of eight dermatologists. Classifiers’ performances were measured in terms of distance from the ideal classifier. Compared with 1- and 2-classifier groups, performances of 3-classifier systems were significantly higher ( P<0.0005 and P<0.001, respectively). No statistically significant differences were found between the 1- and 2-classifier groups ( P=0.352). While the dermatologists group showed a level of performances significantly higher than the 1-classifier systems ( P<0.020), no differences were found between the multiple classifier groups and the dermatologists groups, indicating comparable performances. This work suggests that a suitable combination of different kinds of classifiers can improve the performances of an automatic diagnostic system.