Statistical approaches for typical challenges in medical image analysis

I develop novel methods to tackle challenges in medical imaging, including insufficient number of cases, unbalanced data and poor image quality. My projects focus on computer vision techniques such as active contours, and machine learning approaches such as representation learning, sparse filtering, variational techniques and attention-based approaches. For example, one of my recent work introduces a new framework that consists of 1) triplet loss, 2) perceptual loss and 3) Kullback-Leibler (KL) loss. By using the advantages of both Variational Autoencoders (VAE) and Metric learning, the framework would be capable of doing two tasks at the same time - learning image representations with fine-grained information and doing stochastic inference. My technique supplied much richer image representation, showing a significant improvement of the classification (MNIST and Shoes datasets) over the traditional VAE