We are currently working on pattern classification of DTI-CSM using a machine/deep learning-based approach. Diffusion Tensor Imaging (DTI) in one simple sentence can be introduced as the imaging modality of choice for microstructure visualization. Cervical Spondylotic Myelopathy (CSM) is a degenerative spinal cord disorder in the cervical region (neck). It is the most serious complication of cervical spondylosis and the most common cause of spinal dysfunction especially in patients older than 55 years old. Our research aims to construct a bridge between these two, by proposing a reliable and promising machine/deep learning-based multi-classification method to categorize DT images of patients with/without CSM, by utilizing diffusion tensor scalar parameters and other useful and or accessible features. This will significantly expand the ability of spine surgeons and radiologists to discriminate the degree of spinal cord damage severity, which would have a direct influence on how the surgeon decides about the patient’s treatment type.

Contact: Maryam

Visual details of digital humans in games, VR/AR applications, and films are becoming significantly more demanding. Hair, as a vital component of the human’s appearance, plays an important role in producing digital characters. However, the generation of realistic hairstyles usually needs professional digital artists and complex hardware, and the procedure is often time-consuming. Thus, accurate capture of real-world hairstyles can greatly benefit the production pipeline.

Our research topic is image-based 3D hair model reconstruction using deep learning. It can be divided into two parts: 2D hair analysis and 3D hair strand reconstruction. 2D hair analysis includes 2D hair strand extraction, 2D hair segmentation, hairstyle pattern recognition, and braid structure analysis. The 3D hair strand reconstruction system aims to reconstruct physically-plausible hair models of both unconstrained and constrained hairstyles from a hair image.

Contact: Chao Sun