Meitu Imaging & Vision Lab (MTlab), the AI research and development hub of Meitu Inc., has secured the top place in the Lesion Segmentation Task at the International Skin Imaging Collaboration (ISIC) Challenge 2018. This is the first time that MTlab participated in the ISIC Challenge. As the only team with an accuracy rate of over 80% in the final testing, MTlab bested more than 80 competing teams from around the world. Other competitors included, but were not limited to, participating companies and labs like Tencent's Youtu Lab, and Lenovo Research, as well as academic institutions like Nanyang Technological University, National University of Singapore, and Weill Cornell Medicine.

Meitu's MTlab Won the No.1 Place in the ISIC Challenge 2018

Based on its deep convolutional neural networks, the MTlab team proposed a two-step approach that combines modern detection and segmentation networks. The detection network locates the bounding box of the target lesion and the segmentation network then masks out the area around the lesion. With several notable improvements in training and network architecture, the proposed approach has reached the highest-performing segmentation accuracy to date.

'Meitu aims to help our users become beautiful, not only online, but also in real life. That's why in 2017, MTlab began investing in biometric skincare, utilizing AI technology to provide smarter skin management solutions. The technology used in this competition will be used in future product innovations for skin inspection and care,' said Qian Chengyao from MTlab.

Earlier this year, Meitu launched a skin analysis app and a skin assessing smart hardware device that can assess each user's unique skincare needs. Each user that tried these new products received a personalized beauty plan based on a list of recommended skincare products.

ISIC was founded in 2016 in an international effort to improve the effectiveness of melanoma diagnosis, and is sponsored by the International Society for Digital Imaging of the Skin (ISDIS). ISIC provides various images of skin problems for early detection and the 2018 ISIC Challenge focused on skin lesion analysis for the purpose of melanoma detection. Contestants were asked to use algorithms to create predicted responses.

Skin segmentation, widely used in biometric applications such as face detection, face recognition, face tracking, and hand gesture recognition, has been one of the hottest topics in biometric imaging in recent years. The process is particularly challenging in that lesions may have a wide variety of appearances in terms of size, color, and texture, and may be located on different parts of the anatomy. In other words, there are a variety of places lesions may be found and some cases could even be obscured by body hair.