Algorithm development for wrinkle evaluation based on artificial intelligence (A.I) technology
237
Presented by: Jin-Hee Shin
Wrinkle formation is considered as a major indicator for skin ageing. Although there are the methodological evaluations for preventing wrinkle formation, it is important for clinicians to be able to grade wrinkles objectively. Unfortunately, the results as definitively classified criteria are unclear depending on each researcher due to visualized degree. Recently, artificial intelligence (AI) strategies are providing a potential solution to the growing demand in health care fields for diagnosis and are potent to redefine how researchers can deliver efficacy evaluation methodology to the next generation. In this study, we developed a wrinkle detection algorithm based on a A.I technique. This system can accurately and rapidly detect wrinkles.
Grades classified by visual assessment were used as references and we chose the control-learning method provided with the algorithm (Visual assessment vs A.I evaluation). Five hundred images were submitted to a machine learning algorithm for reading. Acquired images are preprocessed by Face Mesh solution using MediaPipe platform on Google. The process provided 9 ROI from one photograph, and we have acquired consecutively five thousand ROIs based on machine learning. Data augmentation was performed through the image conversion process such as image rotation, brightness and contrast adjustment, and then we analyzed over one hundred thousand images augmented.
Out of 500 volunteers, the pickup rate for major wrinkles was 100%, although it for whole wrinkle grade was approximately 70%.
In this study, we could verify that the performance AI-based wrinkle GRADE is useful and actuatable to evaluate showing efficacy results for preventing wrinkle formation by providing only images. This enables researchers to track the progress of antiwrinkling techniques such as anti-aging cosmetics. The algorithm gives a chance to apply photographs from mobile device, showing the results similar expert grading. Moreover, our system is low cost as the wrinkle detection can be simply based on photographs.
Grades classified by visual assessment were used as references and we chose the control-learning method provided with the algorithm (Visual assessment vs A.I evaluation). Five hundred images were submitted to a machine learning algorithm for reading. Acquired images are preprocessed by Face Mesh solution using MediaPipe platform on Google. The process provided 9 ROI from one photograph, and we have acquired consecutively five thousand ROIs based on machine learning. Data augmentation was performed through the image conversion process such as image rotation, brightness and contrast adjustment, and then we analyzed over one hundred thousand images augmented.
Out of 500 volunteers, the pickup rate for major wrinkles was 100%, although it for whole wrinkle grade was approximately 70%.
In this study, we could verify that the performance AI-based wrinkle GRADE is useful and actuatable to evaluate showing efficacy results for preventing wrinkle formation by providing only images. This enables researchers to track the progress of antiwrinkling techniques such as anti-aging cosmetics. The algorithm gives a chance to apply photographs from mobile device, showing the results similar expert grading. Moreover, our system is low cost as the wrinkle detection can be simply based on photographs.