Segmentation of Medical Images via Deep Learning Techniques: Current State-Of-The-Art and Perspectives
Despite the substantial research interests and efforts devoted to segmentation of medical images, automatizing this task still remains challenging. Clinicians still rely on manual delineations, a prohibitively time-consuming process, which depends on rater variability, leading to substantial inconsistencies in the segmentation. On the other hand, deep learning has emerged as a powerful classification tool that is breaking records in some other domains, such as pattern or speech recognition. Inspired by this, we aim at bringing the power of deep learning techniques to overcome the limitations of classical segmentation methods.
Jose Dolz works as post-doctoral researcher at the Laboratory for Imagery, Vision and Artificial Intelligence (LIVIA) at the ETS in Montreal, Canada. He was awarded by a Marie Curie Actions fellowship which supported his PhD, in France. Recently, he moved to Montreal, where he tries to bring the power of deep learning techniques to the medical field in order to address some of their challenges. Particularly, his current research focuses on the segmentation of medical images at different locations, such as brain, heart or spine, for example. In addition, his research interests also lie on the optimization of regularization models which, combined with deep learning methods, may bring a breakthrough to the field.