Overview
Skin-App denotes an interdisciplinary research strand between HSLU and the University Hospital Zurich that incorporates a broad variety of activities related to automated detection and quantification of hand eczema and other dermatoses. In an initial project a very promising approach to objectively quantify eczematous skin has been evaluated and prototyped based on computer vision and machine learning techniques. The algorithm learns the characteristic structures of eczema skin from high-resolution training images labelled by a professional dermatologist. However, producing training and test data in this way is expensive and does not scale. Moreover, because only a single expert is being involved, the data is statistically biased and as a consequence of the annotation method also partially erroneous. This lack of high-quality, unbiased training and test data makes the accurate statistical evaluation of existing methods difficult and therefore considerably handicaps any further algorithmic progress. As a reaction the project team has constructed an interactive photo-box for patients to automatically take high-resolution photographs with standardized camera setting during hospital consultation. In order to avoid statistical bias, these images must be shown to a representative group of professional dermatologists for labeling eczema skin patches. Once available, the labelings of different experts can be consolidated in order to derive a consensus diagnose for training and evaluation purposes of machine learning classifiers. We design and implement in this project an online platform for harnessing high-quality training and test data based on a consensus diagnose between experts. In combination with the above mentioned photo-box, such a platform will not only flatten the ground for further algorithmic progress in the Skin-App project strand, but it will also enable new forms of clinical studies comparing eczema assessment and scoring between experts and can also be used as an educational platform for students in medical science.
