RESUMO
BACKGROUND: Despite the existence of diverse instruments to assess cellulite, its high prevalence and the continuous advances in its treatment makes the development of new, more objective methods for evaluation necessary. AIM: To study intraobserver validity and reliability (test-retest) of textural analysis using co-occurrence matrices on photographic images in the evaluation of cellulite in a Spanish population and its possible relationship to the degree of cellulite. METHODS: Twenty-seven women were selected for this reliability study (mean age 26.41 SD = 6.16). Digital photographs were taken under standardized conditions in contraction and relaxation of the femoral gluteus region. The areas of interest of the photographs were selected at two different times a month apart. Textural parameters studied were energy (ASM), entropy, contrast, Homogeneity (IDM) and textural correlation. Reliability was analysed by the intraclass correlation coefficient (ICC). Differences between laterality and between contraction and relaxation were performed by analysis of variance for repeated measurements. Correlation between Cellulite Severity Scale (CSS) and the textural parameters by means of the Pearson correlation coefficient was studied. CSS was re-coded to a binary variable, performing a differentiate analysis for each laterality with this variable and the textural parameters. RESULTS: In the intraobserver reliability analysis ICC was high (≥0.80) in seven parameters and excellent (≥0.90) in 35 parameters. In general, CSS and textural parameters showed more cellulite severity in right areas than in left ones. Correlation coefficients showed a moderate correlation between textural parameters and CSS score. The multivariate discriminant model obtained with textural parameters classified a high percentage of images (96% right side and 82% left side). CONCLUSION: Textural analysis used to assess cellulite on the backs of thighs and buttocks proved to be an instrument that has excellent reliability, moderate correlation with CSS and a satisfactory power classification model.
