Resume of Dr. Juan Felipe Pérez-Juste Abascal

Short biography

I am Theoretical Physicist from University Autonoma of Madrid, with an M.Sc. degree in Applied Mathematics from The University of Manchester and a Ph.D. degree in Medical Physics/Computer Science from University College London. After the Ph.D., I held several postdoctoral positions, with Centrale Supelec, Hospital General Universitario Gregorio Marañon and University Carlos III of Madrid. In 2016, I gained a Marie Skłodowska-Curie fellowship to come to CREATIS lab, Lyon. In industry, I was an R&D engineer at Alten, Madrid, and at Carestream dental, Paris.

My contributions lie in the fields of inverse problems, optimization, tomographic reconstruction image restoration and deep learning, with a particular focus on medical applications (MRI, CT, spectral CT, diffuse optical tomography, electrical impedance tomography, PET).

Lately, I have been interested in computer vision applications, such as object detection, video processing and surveillance applications, and in NLP and LLMs.

Currently, I am working as a consultant Freelancer where I could continue my contribution and passion on imaging, applied maths and AI.

Long biography : Projects and contributions

My passion for research and imaging was born in 2002-2003 during my master thesis at the School of Mathematics at The University of Manchester, where I worked on uniqueness for Electrical Impedance Tomography (EIT) under the supervision of Bill Lionheart.

After the master, I pursued a PhD thesis in the EIT group at the Medical Physics Department at University College London (UCL), a multidisciplinary group led by David Holder and that has been pioneer in the use of EIT for imaging brain function (2004-2007). The thesis aimed at developing algorithms for EIT of brain function and was co-supervised by Simon Arridge and Richard Bayford. It resulted in five peer-reviewed publications on both theoretical/modelling and experimental work. The experimental work optimized reconstruction of neonatal functional data and led to data preprocessing using principal component analysis, modelling of the covariance of the data for the maximum a posteriori estimator and optimal selection of the regularization parameter. The technical contributions investigated EIT for anisotropic media and provided a FEM solver, a realistic anisotropic model of the human head and a novel uniqueness result for reconstructing the anisotropic conductivity tensor with constraints. The validation of the computer-simulation tool was selected by The Institute Of Physics as highly innovative (Phys Meas 2008), the human model work was published in Neuroimage and had been highly cited (Neuroimage 2008, 136 citations, GoogleScholar), and the uniqueness result was published in Inverse Problems.

During a postdoc at Centrale Supelec in 2007-2008, I worked on Non-Destructive Testing (NdT) of metal tubes using Eddy-current imaging and level sets, supervised by Dominique Lesselier and in collaboration with Oliver Dorn. Contributions included modeling based on a vector domain integral formulation and a shape reconstruction algorithm using level sets.

At different occasions, I have been involved in NdT using electrical capacitance tomography in collaboration with Manuch Soleimani, from the University of Bath.

In 2008, I joined Alten Engineering, Madrid, as an R&D engineer, to model a competitive electricity market based on a game-theoretic approach with linear programming and seasonal forecasting.

In 2009, I joined the Laboratory of Medical Imaging (LIM), a multi-disciplinary group based at Hospital Geneneral Universitario Gregorio Marañón (HGUGM) and led by Manuel Desco. Within the framework of the EU FP7 “FMT–XCT system” project, led by Juan Jose Vaquero at LIM, I contributed to the development of a hybrid Fluorescence Molecular Tomography and micro-CT system for preclinical applications. This work had major contributions on both modelling the forward solution and on image reconstruction, in collaboration with Jorge Ripoll and Simon Arridge. I was glad to collaborate with Juan Aguirre, Judit Chamorro-Servent and Teresa Correia. Contributions included studying the influence of absorption and scattering coefficients on quantification (Green functions, FEM) and optimal reconstruction for large-scale problems, using the split Bregman approach with a non-negativity constraint and incorporation of the knowledge of an anatomical XCT prior image into the reconstruction.

From 2011 to 2016, at LIM and at University Carlos III of Madrid (UC3M), I was the technical leader of a new research-line dedicated to provide solutions to specific imaging problems based on the Compressed Sensing (CS) method. Applications were driven by clinical needs at HGUGM. We tackled the following challenging problems: Accelerating MRI acquisitions, low-dose imaging, motion-artefact reduction and limited-view imaging. A wide range of applications were investigated during this period, the large majority were imaging problems with high dimensional data, as these had higher potential of improvement with the CS methodology, in cardiac-cine MRI, in fMRI, in diffusion MRI, in respiratory Gated CT, in limited view CT, and in dynamic PET. We obtained high acceleration factors (x8-x15) and up to six-fold decrease in dose. Together with Manuel Desco, I co-supervised the PhD theses of Paula Montesinos on cardiac-cine MRI and Cristina Chavarrias on fMRI. I was glad to collaborate with Monica Abella and Alejandro Sisniega-Crespo on CT, Eduardo Lage on PET, and Juan Parra on lung diffusion MRI with hyperpolarised gases.

In 2016, I was granted an EU H2020 Marie Skłodowska-Curie fellowship (project “SUCCESS”) to come to CREATIS lab, Lyon, under the supervision of Francoise Peyrin, with the goal to investigate and develop new image reconstruction algorithms for the new generations of spectral CT scanners. This work lied also within the framework of another two projects. The first one, the EU H2020 “SPCCT Project” led by Philippe Douek, Lyon, and Philips Research Laboratories, Hamburg, Germany, which aimed at developing an unique spectral CT scanner and contrast agents for cardiovascular diseases. Our contributions comprised of a theoretical proof of non-convexity and feasible material decomposition algorithms based on optimization approaches and deep learning.

The second project -- ANR “SALTO” – aimed at early diagnosis of Osteoarthritis using spectral CT and was led by Francoise Peyrin in collaboration with Cristine Chappard, from B2OA, Paris. We showed for the first time that direct visualization of both cartilage and bone impairment is possible with spectral CT and that “virtual monoenergetic images” improved image quality with respect to standard CT. For this, we were granted two Beam Time Projects at The European Synchrotron Radiation Facility (ESRF) for the validation of spectral CT in comparison with monochromatic synchrotron radiation CT. Together with Francoise Peyrin and Nicolas Ducros, I co-supervised the PhD thesis of Suzanne Bussod on deep learning for spectral CT. We also contributed to the PhD thesis of Valeriya Pronina, from Skolkovo Institute of Science and Technology, Moscow, on learnable algorithms for image restoration.

Since 2017, following the AI revolution, I focused on deep learning for inverse problems and medical imaging. A deep learning based material decomposition method that led to a large improvement in image quality with respect to more classical approaches on human data. Exploiting transfer learning, we learnt the detector response function, and following a Sim2Real approach, we reconstructed human SCT data where very few datasets were available.

Following the success of compressed sensing on high-dimensional data, we developed corresponding deep learning approaches. We leverage a prior image and efficient data pipelines for 3D denoising of SCT images, and we leverage higher-order convolution layers for denoising of single-pixel hyperspectral images and for motion-artefact correction.

Combining iterative optimization algorithms and learning approaches, we investigated and compared post-processing methods, plug-and-play priors and unrolled methods for CT reconstruction.

At CREATIS, I was one of the creators and organizers (2018-2020) of the monthly meeting “Coffee Machine Learning” that gathered lovers and followers of machine learning across all lab departments. The meeting was very successful and is still on-going.

From 2021 to mid-2022, I joined Carestream dental as a R&D engineer to work on image processing, image reconstruction and deep learning.

Lately, I have been interested in computer vision applications, such as object detection, video processing and surveillance applications, and in NLP and LLMs.

Currently, I am working as a consultant Freelancer where I could continue my contribution and passion on imaging, applied maths and AI.