Clinical-view versus ELM: an investigation into image types in the context of skin lesion screening

Abstract

Melanoma, the most serious form of skin cancer, is increasing in incidence in countries with predominantly white skinned populations. Automated tools have been proposed to help detect this most visible of cancers. Current automated systems for detecting melanoma analyse images of skin lesions for relevant image features, and classify the images based on those features. There are two types of image available to be used in such systems, Clinical-view or Epiluminescent microscopy (ELM) images. ELM images reportedly allow more accurate assessment of skin lesions in the clinical setting, but this finding has not been proven in the context of an automated system.

This research has evaluated the question of Clinical-view versus ELM images in an automated screening system. Two methods of implementing a screening system were considered in this research. Firstly, the ‘diagnosis system’, which is based on previous work in this field, and secondly, a ‘dermatologist assessment system’, which is an original method of implementing an automated screening system. The Clinical-view versus ELM question was considered for both of these systems.

Specifically, two automated systems were developed. The first analysed Clinical-view images, while the second processed ELM images. From the analysis, each system attempted to classify lesion images into two groups. For the diagnosis problem, the lesion was either ‘melanoma’ or ‘benign’. For the ‘dermatologist assessment’ problem, the groups were ‘excised’ or ‘not excised’. The results raise doubts over the current emphasis on ELM images in the automated diagnosis case. Similarly, it appears that Clinical-view images are of more use for reproducing ‘dermatologists assessment’. We have also shown that the ‘dermatologist assessment’ approach to screening skin lesions is a viable and potentially useful alternative to the current emphasis on the diagnosis approach.