Bayesian decision fusion for enhancing melanoma recognition accuracy

Abstract

This paper studies the problem of automated noninvasive skin cancer (melanoma) detection from digital images of skin lesions. It proposes the use of Bayesian Decision Fusion of a multiple of classifiers to enhance the melanoma detection rates. A comparison with other decision fusion systems along with standalone classifiers in terms of accuracy and confidence intervals is made. The relation between confidence distribution and accuracy over different classification systems is studied. Performance evaluations of the proposed Bayesian Decision fusion method shows that it results in improved recognition accuracy compared to standalone Skin Lesion classifiers. It also provides comparable confidence intervals and can offer stable recognition rate. Hence, it can lead to increased chances of non-invasive melanoma detection.