Uma abordagem estatística para diagnóstico e classificação do fenômeno de Raynaud a partir de dados termográficos

Abstract

Raynaud’s phenomenon (PR) comes in two forms, primary and secondary. The diagnosis of the condition is of paramount importance, as is the differentiation between its two forms.Clinically, this differentiation is not possible, and thermography emerges as a complementary diagnostic tool to clinical examinations, being a low cost, noninvasive and easily operable technique. Thus, the aim of this study is to verify the feasibility of using thermographic images to assist in the diagnosis of patients with RP through statistical analyzes such as Principal Component Analysis (PCA) and Geostatistics. Data collection was performed in 279 individuals, 255 of these were undiagnosed individuals, being other individuals diagnosed with Raynaud’s Phenomenon in its primary form (RP1) and 13 with the secondary form (RP2). The collection environment was maintained at 23◦C, using a thermal camera (Fluke Ti400) to monitor the reheat pattern after a cold stress protocol, i.e., under hand immersion to the carpal level in a container with water at 10◦C for 60 seconds. Images were recorded at the following times: before immersion, immediately after (t = 0), and thereafter every 5 minutes, until 20 minutes after immersion (t = 5, t = 10, t = 15 and t = 20). For descriptive and exploratory analyzes, composed by ANOVA analysis and Tukey test, 36 individuals were selected, 11 previously classified with primary Raynaud’s Phenomenon (RP1), 13 classified with the secondary phenomenon (RP2), and 12 healthy (S-control group). It was found that there is no significant difference between the averages and variances of temperatures. However, when considering the differences between the temperatures of both hands of each individual, there is significant difference in the variability of the RP2 group for the others. An analysis of variance (ANOVA) with Tukey test revealed that the groups RP1 and RP2 have very similar behaviors, but with RP1 presenting faster balance than RP2, and both differing considerably from healthy cases. After performing the Principal Component Analysis to reduce the dimensionality of the data, considering the 279 individuals, it proved ineffective to distinguish the groups, and it revealed that this classification involves a space-temporality to be taken into account. Finally, geostatistical analyzes were performed on the 36 selected individuals. Mathematical models were found that best represent each individual in each time, and krigings of each image were generated for a visual validation of the found model. The models fit the data faithfully and provided good representations, being the most frequent models: Bessel, Exponential and Gaussian.