Research Synopsis

My research at CNRS (LaBRI UMR 5800, University de Bordeaux) on medical image processing covers different fields, from preprocessing to automatic detection of diseases. My work mainly focuses on four topics: Deep learning for medical imaging, BigData analysis for medical imaging, quantitative MR analysis, computer-aided diagnosis and image enhancement.

Deep Learning for Medical Imaging

In the past years, Deep Learning (DL) methods have a great success in computer vision. However, medical imaging requires specific adaptations to take benefit from the full capability of such technics. Recently, we started integrating DL frameworks into our tools to address different chalenges such as image enhancement, registration, segmentation and computer-aided diagnosis.

In this field our main contributions are: i) the development of new DL methods for denoising, ii ) the development of new DL methods for registration and iii) he development of new DL methods for MRI segmentation.

BigData for Medical Imaging

There is no consensus in literature about lifespan brain maturation and senescence, mainly because previous lifespan studies have been performed on restricted age periods and/or with a limited number of scans, making results instable and their comparison very difficult. Moreover, the use of non-harmonized tools and different volumetric measurements lead to a great discrepancy in reported results. Thanks to the new paradigm of BigData sharing in neuroimaging and the last advances in image processing enabling to process baby as well as elderly scans with the same tool, new insights on brain maturation and aging can be obtained.

In this field our main contributions are: i) the development of new tools able to robustly process massive datasets, ii ) the first analysis of brain structure volume over the entire lifespan and iii) its application to Alzheimer’s Disease.

Cloud solution for Medical Image Analysis

Most of the developed pipelines for MR image analysis are packages that need to be downloaded, installed and configured. Installation step can be complicated and thus may require an experimented person not always available in a research laboratory or clinical context. In addition, users have to be trained to use the software and computational resources have to be allocated to run it. We have tried to overcome all these problems by deploying our volBrain pipeline through a web interface providing not only access to the software but also sharing the computational resources of our institutions. Thus, using the volBrain pipeline does not require any installation, configuration or training. The volBrain platform works remotely through a web interface based on a SaaS (Software as a Service) model to automatically provide a report containing volumetric information.

In this field our main contributions are: i) development of the volBrain platform, ii) the integration of several pipeline covering a large range of neuroimaging applications, and iii) the processing of almost 100.000 MRI for worldwide users

Quantitative MR Analysis

Magnetic resonance (MR) imaging plays a crucial role in the detection of pathology, the study of brain organization, and clinical research. Every day, a vast amount of data is produced in clinical settings, preventing the use of manual approaches for data analysis. Consequently, the development of accurate, robust, and reliable segmentation techniques for the automatic extraction of anatomical structures is becoming an important challenge in quantitative MR analysis.

In this field our main contributions are: i) the proposition of a new segmentation method based on patch-based estimator, ii) its validation on many cerebral structures from brain to deep brain structures, and iii) its application to quantitative analysis in the context of different pathologies such as Alzheimer’s Disease, Parkinson’s Disease and Infant Autism.

Computer-Aided Diagnosis

In last decade, computer-aided diagnosis was a rapidly growing field of research. This field aims to develop new image analysis techniques in order to assist doctor in image interpretation. Usually based on automatic pipeline involving preprocessing, feature extraction and feature classification; CAD can be dedicated to pathology detection or to predict its evolution. In addition, for some tasks, such as the simultaneous comparison of a large number of images or the detection of subtle anatomical changes caused by diseases, computer is now an essential tool.

In this field our main contributions are: i) the invention of a new scoring method (SNIPE) for brain morphometry analysis, ii) the proposition of a robust framework for patient's classification based on SNIPE, and iii) its validation for automatic and accurate detection of Alzheimer's Disease.

Image Enhancement

In medical imaging, image enhancement is an important preprocessing step used to improve image analysis. Since the noise and the image resolution directly impacts the accuracy of automatic approaches such as segmentation and registration, it is important to improve the quality of images used during quantitative analysis.

In this field our main contributions are: i) the introduction of the nonlocal means filter for MRI processing, ii) its adaptation to DWI, ultrasound and multi-photon images, and iii) the investigation of its impact on tissue classification and cortical surface extraction.