Multiset multitemporal canonical analysis of psoriasis images

Nowadays, the medical tracking of dermatological diseases is imprecise, mainly due to the lack of suitable objective methods to evaluate the lesion. The severity of the disease is currently scored by doctors merely by means of visual examination. In this work, multiset canonical correlation analysis over registered images is proposed to track the evolution of the disease automatically. This method transforms the original images into sets of variables that exhibit, decreasing degree of similarity, based on correlation measures. Due to this property, these new variables are more suitable to detect where changes occur. An experiment with 5 different time series collected from psoriasis patients during 4 different sessions is conducted. The analysis of the obtained results points out some patterns that can be used both to interpret and summarize the evolution of the lesion and to achieve a better image registration.