Current research interests
The common thread of my research is to devise new mathematical founded computational tools for analyzing medical images acquired through different imaging modalities. Those tools can also be used for improving the acquisition of images. Efficiency is of utmost importance, provided that the ultimate goal of these tools is to support physicians in clinical practice. I am very interested in bringing these new developments from basic research to final products that can be used in clinics, for which collaborations with the industry is a must. In particular, I am interested in using artificial intelligence to solve image analysis problems, the modeling of perception theories into computational methods, the use of tensor-based image processing, and the optimization of image acquisition in advanced MRI imaging.
1. Analysis of Brain Connectivity
Diffusion magnetic resonance imaging (dMRI) is an MRI technique that can locally detect anisotropies in the movement of water molecules in the brain. Such anisotropies are used by tractorgraphy algorithms for estimating the most likely paths followed by the neural tracts in the white matter of the brain. Although the current tractography methods have good sensitivity for extracting the main neural fiber bundles from dMRI data, they also have very low specificity, which hinders its use in research and clinical environments.
Functional MRI (fMRI) is a magnetic resonance imaging (MRI) technique that can be used to detect neural activation in the brain. In resting state fMRI (rs-fMRI) no stimuli is shown to the subjects. This imaging technique can be used to study functional connections between brain regions. rs-fMRI can be used to detect connectivity impairments due to diseases. rs-fMRI can also be used to analyze the dynamic connectivity of the brain, which can be used to determine causality among activation patterns. Analysis of rs-fMRI data is challenging due to it is prone to many artifacts, especially in acquisition of animal data.
Our research aims at devising artificial intelligence-based methods for analyzing both dMRI and rs-fMRI data and the combination of them, with focus on human and small animal data.
2. Magnetic resonance elastography
This project aims at characterizing the mechanical properties of the brain through Magnetic resonance elastography (MRE) in adolescents and older children and at correlating them with risk factors for developing anxiety disorders. Our long-term goal is to characterize the mechanical properties of the brain at different ages with the aim of improving the diagnosis and treatment of anxiety disorders in the young population. Our long term aim is to extend this project to characterize the mechanical properties of the brain at different stages of brain development, which can be used to improve the diagnoses of various neuropsychological diseases, especially at early stages where treatments are more likely to have an effect, and to better track the response of patients to treatments.
3. Synthetic MRI aging
One challenge in image analysis of Alzheimer’s patients is to disentangle the effects of aging from the disease itself. For this, it is necessary to use a large number of images of healthy subjects at every specific age, which is not available in the current longitudinal imaging datasets. In this project, we will develop an AI-based solution to generate magnetic resonance images (MRI) at different ages by simulating the changes in the images due to aging. In particular, given a structural MRI acquired at a specific age, the solution will be able to generate MRI of the same subject at older and/or younger ages. AI-based diffeomorphic registration will be used extensively in this project since, unlike standard generative models, it is able to preserve the anatomy of the brain.
We will validate the method for mild cognitive impairment (MCI) by comparing the subjects' actual time course with the ones of simulated healthy and MCI subjects. Having a tool to accurately generate new time points of MRI data will have a big impact on the diagnostic work-up by helping to predict clinical progression of patients over time (prospective), or providing information about premorbid states (retrospective).
1. Deep Segment
In this project, we aim at developing a workflow system for the complete structure segmentation process in medical and biological imaging covering all parts of the process: sample preparation, scanning, both manual and automatic segmentation tools, analysis and visualization. The system will cover both conventional methods as well as deep learning-based approaches. It will be designed for large studies and capable of handling also very large images with a strong focus on quality, usability and efficiency. This project is funded by Eurostars - Vinnova grant No. E!11626 1 949 648 SEK, 30 months from 2017-2020. P.I. Rodrigo Moreno.
2. Biomechanics at interactive speed
Current pipelines for using biomechanical information in clinics are computationally expensive. In some cases, the calculations can take days or weeks in large computational infrastructures. The final output is then condensed to a single image, or even a few values, indicating for example the risk of fracture or the bone strength. AI may open the door to a new practice where doctors can get the biomechanical parameters interactively. Our research aims at developing an AI-based approach for biomechanics, that can avoid the expensive simulations. The goal is to bring the process to interactive speed, which is required in clinics.
3. Analysis of trabecular bone microstructure
The economic burden of osteoporosis-related factures on public health systems is huge, and it is actually more severe in Nordic countries. Osteoporosis is a disease that weakens the internal structure of trabecular bone. Thus, it is of utmost importance to devise quantitative assessment tools for analyzing 3D images of trabecular bone. The emergence of high resolution imaging modalities, especially Cone Beam Computed Tomography (CBCT) and High Resolution Peripheral Quantitative Computed Tomography (HR-pQCT) has opened the door to perform such analyses in vivo. We have developed in the last few years several methods for analyzing the microstructure of trabecular bone in gray-scale, including trabecular thickness quantitation, classification of trabeculae and estimation of different types of fabric tensors. The methods have already been validated by comparing them to an established reference method (micro-CT) in bone specimens. However, before these methods are used in clinical routine, it is necessary to perform an extensive validation, which is one of the objectives of this project.
We plan to carry out clinical studies that aim at understanding the factors involved in the local and global changes of the structure and mechanical competence of trabecular bone due to osteoporosis. This is expected to be clinically useful in particular in the evaluation of new drugs. This project is funded by Eurostars - Vinnova grant No. E!9126 1 543 354 SEK, 26 months from 2015-2017.
4. Analysis of blood vessels
Atherosclerosis is one of the leading causes of death in Western countries. Computed Tomography Angiography (CTA) and Magnetic Resonance Angiography (MRA) have made possible to quantify the stenosis degree noninvasively in coronary arteries. However, factors such as low resolution, noise and confounding surrounding tissues makes challenging the automatic analysis of coronary arteries. In this line, one of the aims of the project is to develop more robust methods for analyzing coronary arteries. A second aim is to characterize the composition of atherosclerotic plaques. Besides CTA and MRA, advanced MRI sequences (Dixon and PC-MRI) can be used to characterize the composition of atherosclerotic plaques in vivo. Validation of methods through histopathology is a necessary step towards its use in clinical practice, which is also an objective of this project.
5. Medical image analysis through tensor voting
Tensor voting is a versatile technique that aims at extracting salient information from noisy images by means of perception-based rules using tensors. While tensor voting has extensively been applied to different problems in computer vision, its use in medical image analysis applications has hitherto been scarce. The main reason of this is that appropriate extensions of tensor voting are necessary for applying this technique to different medical imaging modalities, especially Computed Tomography (CT) and different pulse sequences of Magnetic Resonance Imaging (MRI). These extensions can potentially be used in different applications, such as detection of bifurcations in blood vessels, detection of vortices in blood flow, and detection of nodes in trabecular bone, among many others. Thus, the main goal of this project is to propose such extensions. This project is funded by The Swedish Research Council (VR) grant No. 2012-3512, 2 400 000 SEK, 3 years from 2013-2015, P.I. Rodrigo Moreno.
Source Code (available on request)
- Efficient Tensor Voting
- Generalized Mean Intercept Length Tensor
- Classification of Trabeculae
- Distance between sets of points
- Vesselness using the Ring Pattern Detector
3D Image Reconstruction and Analysis in Medicine (HL2027), examiner, course responsible | Course web
Magnetic Resonance Imaging basic principles (CM208V), examiner, course responsible | Course web
Postprocessing in Magnetic Resonance Imaging (CM210V), examiner, course responsible | Course web