On the Usefulness of Pre-Processing Step in Melanoma Detection Using Multiple Instance Learning

Although skin cancers, and melanoma in particular, are characterized by a high mortality rate, on the other hand they can be effectively treated when the diagnosis is made at the initial stages. The research in this field is attempting to design systems aimed at automatically detecting melanomas on the basis of dermoscopic images. The interest is also motivated by the opportunity to implement solutions that favor self-diagnosis in the population. Determining effective detection methods to reduce the error rate in diagnosis is a crucial challenge. Computer Vision Systems are characterized by several basic steps. Pre-processing is the first phase and plays the fundamental role to improve the image quality by eliminating noises and irrelevant parts from the background of the skin. In [1] we presented an application to image classification of a Multiple Instance Learning approach (MIL), with the aim to discriminate between positive and negative images. In [3] we subsequently applied this method to clinical data consisting of non-pre-processed melanoma dermoscopic images. In [2] we also investigated some pre-processing techniques useful for automatic analysis of melanoma images. In this work we propose to use, after applying a pre-processing step, the MIL approach presented in [1] on the same melanoma data set adopted in [3]. The preliminary results appear promising for defining automatic systems that act as a “filter” mechanism to support physicians in detecting melanomas cancer.