Gabriel Bernardino

WARNING: This webpage is under construction

I am a researcher in medical image analysis working at the intersection of applied mathematics, machine learning and image processing. I develop computational techniques for better assessment of the cardiovascular system. My methodological contributions consider the physiology of the underlying clinical problems. I have a strong technical expertise in machine learning, especially in its applications to medical image analysis, and a deep knowledge of cardiac physiology and its biophysical modelling. My view is that physiological knowledge (physical laws) needs to be combined with data-based models to overcome the lack of data quantity in quality in medical imaging, as well as to obtain interpretable models.

Since September 2022, I am a Margarita Salas fellows developping machine learning methods to detect illnesses in fetal echocardiographies, in a collaboration between the clinical research group BCNatal (IDIBAPS) and the Universitat Pompeu Fabra. The objective is to overcome the limiting factors for the application of machine learning for the automatic assessment of fetal images: 1) data scarcity, 2) the high noise in fetal echocardiographic images 3) complexity of the still incomplete fetal heart.

Previously, I was a postdoctoral researcher at CREATIS (Lyon, France), one of the largest public medical imaging research laboratories. There, I worked on exploiting hierarchical relationships between the different imaging modalities to obtain optimal and cost-effective (financial as well as patient safety) data-integration strategies. I proposed a reinforcement learning framework that interactively selects the most relevant modalities during diagnosis. I also worked on hierarchical unsupervised representation learning of multiple modalities using Gaussian Processes.

I started my PhD in 2016 within the MSCA project “Cardiofunxion”, a collaboration between Philips (Paris) and the Universitat Pompeu Fabra (UPF, Barcelona). The PhD was awarded in 2019, with the Cum Laude mention. During my PhD, I developed novel statistical shape analysis and mesh processing methods to identify and understand the effect of altered working conditions on the cardiac shape from medical images (MRI, 3D echocardiography). In my one-and-a-half years stay at Philips, I worked in an industrial setting, integrating my pipelines with their clinical software. The project involved collaboration with renowned clinical research centres specialised in sports cardiology (CHU Caen, Hospital Clinic de Barcelona); and fetal cardiac pathologies (BCNatal).