Algoritmo para detecção, classificação e quantificação de rugas em imagens ópticas ampliadas

Abstract

The analysis of skin relief is crucial in the development of new skin care products, as well as in the evaluation of dermatological treatments. The analysis can be performed by qualitative or quantitative methods. In the present work, it is proposed the development of a new algorithm for classification, detection, and quantification of wrinkles through the image processing from a digital dermatoscope. Two clinical studies were carried out, one with 90 research participants and another with 33 research participants, in which images were collected with the dermatoscope and PRIMOS® equipment for the evaluation of wrinkles. In the first study, images were collected to identify characteristics of the images, for subsequent separation into groups according to the different degrees of wrinkles. In the second study, images were collected at two different times: Day 0 (D0) and after 45 days (D45) using a dermo cosmetic product. A separation of groups was then performed on the set of images collected in the first study, in which the training of a convolutional neural network was applied to evaluate the images, the accuracy of the neural network was 78.5%. Subsequently, a new algorithm was developed to detect wrinkles in the images acquired in the second study, through the application of filters and image transformations that generate a segmented image highlighting the wrinkles. From the pixels belonging to the wrinkles, a roughness calculation method is proposed. The correlation between the values obtained by the PRIMOS® equipment and the proposed system was verified. No correlation was found for the data obtained at D0, however, there was a correlation at time D45 by Spearman's similarity coefficient. When comparing the roughness between times D0 and D45, the treatment was statistically significant both for PRIMOS® and for the proposed methodology data. The wrinkle detection algorithm, in addition to the roughness calculation, demonstrated sensitivity comparable to the PRIMOS® system in evaluating the efficacy of the dermo cosmetic treatment, identifying differences between treatments and the convolutional neural network was able to classify wrinkles. Considering the simplicity of the dermoscope design, compared to other established devices such as the PRIMOS®, the proposed system is promising as an alternative for dermatological evaluations.