Analyse-Images-Medicales |

New Method CMR-Guided Endomyocardial Biopsy in Suspicion Context of Isolated Cardiac Sarcoidosis

Sun, 01 Sep 2024 00:00:00 +0000

Segmentation of cardiac infarction in delayed-enhancement MRI using probability map and transformers-based neural networks

Fri, 01 Dec 2023 00:00:00 +0000

Transformers-Based Neural Network for Cardiac Infarction Segmentation in Delayed-Enhancement MRI

Mon, 16 Oct 2023 00:00:00 +0000

Light Random Regression Forests for automatic multi-organ localization in CT images

Tue, 18 Apr 2017 00:00:00 +0000

Automatically locating organs in CT images

Fri, 30 Sep 2016 00:00:00 +0000

My first turn toward machine learning — taken, with this student, two years before the deep-learning boom.

In the continuation of the project, I co-supervised with Emmanuel Promayon the PhD of Prasad Samarakoon, defended on 30 September 2016 at Université Grenoble Alpes (funded by the French ANR TecSan project “Robacus”). This is the project that tipped me toward machine learning applied to medical imaging. We started two years before the rise of deep-learning segmentation: so we bet not on deep neural networks, but on decision-tree forests (random forests) — a clear-eyed choice for the time, and a formative one.

What this thesis really gave me was not one more method: it was an early familiarity with the strengths and the limits of learning-based approaches — first among them their surprising robustness.

The challenge: locating, not segmenting

To plan a puncture assisted by the LPR robot, one must first locate, in the CT scan, the target organs and those to avoid — a step still done by hand by the clinician, tedious and costly in expert time. Rather than aiming straight for full segmentation (delineating every contour), we tackled the more tractable and equally useful problem of localization: enclosing each organ in a bounding box, automatically.

The contribution

Beyond a thorough analysis of the method, the thesis produced two contributions of real practical reach:

the Light Random Regression Forests: a faster and far more memory-efficient model, at equivalent accuracy — and therefore easier to embed and deploy;
an automatic parametrization that removes settings previously fixed “by hand”, making the method more robust and more reproducible from one dataset to another.

The regression-forest pipeline: training-set preparation, preprocessing, training, then prediction (data in yellow, algorithmic steps in blue)

Robustness — and bias: the phantom-kidney story

One experiment remains, for me, the perfect illustration of what these methods really are. To learn to locate the kidneys, our training database — built from experts’ manual delineations (ours and those of the team’s PhD students) — contained only patients with two kidneys. During the testing phase, the radiologist handed us the image of a patient who had only one. The algorithm dutifully found… two bounding boxes, placing a “phantom kidney” exactly where statistics expected it.

Even then, we were alert to a truth that has lost none of its relevance: a model is only the reflection of the data it is shown. The quality and representativeness of the training set matter as much as the algorithm itself.

What this project represented

With the rise of deep learning, this forest-based localization strategy was later set aside — but it was decisive. It let me approach the machine-learning turn from the foundations, at a time when one still took the time to understand why a method works, what it guarantees, and where it fails. That perspective — robustness, generalization, vigilance about the data — is exactly what I bring today to medical-application prototyping.

Skills involved: doctoral co-supervision · machine learning applied to medical imaging · design of robust, automatic methods · building and critically assessing training datasets.

2017. Samarakoon Prasad N, Promayon Emmanuel, Fouard Céline — Light Random Regression Forests for automatic multi-organ localization in CT images. 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017)
2014. Saramakoon Prasad, Promayon Emmanuel, Fouard Céline — Fully Automatic Organ Localization in Medical Images Using Improved Random Regression Forests. Proceedings of Surgetica 2014

Segmentation, separation and pose estimation of prostate brachytherapy seeds in CT images

Fri, 06 Mar 2015 00:00:00 +0000

Impact de l’orientation des grains iode 125 dans l’évaluation de la dosimétrie à 1 mois d’une curiethérapie prostatique

Wed, 01 Oct 2014 00:00:00 +0000

Medical Image Analysis

Wed, 01 Oct 2014 00:00:00 +0000

When I joined the research group at TIMC laboratory in 2006, I contributed my expertise in medical image processing and analysis to three ongoing applied projects led by my colleagues. These collaborations allowed me to tackle diverse clinical challenges — from real-time surgical vision to oncological dose computation — while immersing myself in the tools and culture of the team.

These three projects are grounded in rigorous academic knowledge production: each contribution led to publications in international peer-reviewed journals and conferences, or to a patent filing. The skills described here were forged alongside that scientific output — they are its direct product.

These three projects embody a conviction that has shaped my entire approach: a medical image is only as good as its usability by the clinician, at the right moment and in the right context.

Prostate Brachytherapy — Orientation of Iodine-125 Seeds

Project led by Jocelyne Troccaz (TIMC-GMCAO)
Co-supervision of post-doctoral researcher Giao Nguyen (50%, with Jocelyne Troccaz)

Context

Transperineal implantation guided by ultrasound — clinical context

Prostate brachytherapy involves implanting radioactive iodine-125 seeds directly into the prostate gland. Calculating the radiation dose absorbed by tumour tissue — dosimetry — is critical for evaluating treatment effectiveness and anticipating side effects. At the time, dosimetry tools modelled seeds as isotropic point sources, ignoring their actual orientation within the tissue.

The research question: does seed orientation influence dosimetry enough to warrant its inclusion in clinical software?

Contributions

To answer this, we first needed to automatically detect and localise seeds in post-operative CT images — a challenging task, as seeds are small, closely spaced, and may overlap in projection.

Implanted seeds as visible in a CT slice: single seeds, clustered seeds, and pelvic bones

We developed a complete pipeline for:

Segmentation of iodine seeds in CT images
Separation of seeds that touch or overlap
Pose estimation (3D position + orientation) of each individual seed

3D localisation of brachytherapy seeds in the prostate (CamiTK visualisation)

Results confirmed that seed orientation had a measurable impact on dosimetry — a clinically significant finding for post-operative evaluation and treatment follow-up.

Skills Applied

3D medical image processing (CT)
Geometric modelling of radioactive sources
Software development and integration into the CamiTK platform for clinical use
Validation protocols on real patient data

[IEEE TBME 2015] H.-G. Nguyen, C. Fouard, J. Troccaz — Segmentation, separation and pose estimation of prostate brachytherapy seeds in CT images —
[Cancer/Radiothérapie 2014] F. Meneu, H. Nguyen, C. Fouard et al. — Impact de l’orientation des grains iode 125 dans l’évaluation de la dosimétrie à 1 mois —
[Physica Medica 2013] F. Meneu, G. Nguyen et al. — Consideration of seeds orientation in prostate brachytherapy and dosimetry analysis —

3D Ultrasound-Based Bone Tracking

Co-supervision at 50%, with Jocelyne Troccaz
PhD thesis of Jonathan Schers (2006–2009)

Ultrasound-guided orthopaedic intervention: the clinical context

Context

Orthopaedic procedures — prosthesis placement, osteotomies — require real-time tracking of bone structures during surgery. Classical techniques rely either on X-ray imaging (radiation) or on invasive bone markers. The goal of this project was to explore a minimally invasive alternative: tracking bone structures using 3D ultrasound.

Contributions

We proposed a per-operative tracking method based on 3D/3D ultrasound image registration: from a reference volume acquired at the start of the procedure, bone displacements are tracked by registering each new acquisition against that reference.

Segmentation of the bone surface in an ultrasound image

Key scientific contributions:

Segmentation of bone surfaces in ultrasound images (noisy signal, reflection artefacts)
Rigid and deformable registration methods adapted to the ultrasound modality
Panoramic reconstruction from multiple partial acquisitions
Evaluation on cadaveric anatomical specimens

Skills Applied

Ultrasound signal processing
Multimodal medical image registration
Design of a ground-truth-free registration validation method
Doctoral co-supervision (50%)

[IUS 1007] J. Schers, J. Troccaz, V. Daanen, C. Fouard, C. Plaskos, P. Kilian - 3D/4D ultrasound registration of bone -
[IJCARS 2009] J. Schers, J. Troccaz, C. Fouard, C. Plaskos, O. Palombi — 3D/3D ultrasound registration for panoramic volume reconstruction —
[CAOS 2010] J. Schers, C. Fouard, J. Troccaz — Non invasive ultrasound-based bone tracking —

Distributed Laparoscopy — Extended View of the Surgical Field

Co-supervision at 50%, with Philippe Cinquin
PhD thesis of Christophe Boschet (2007–2010)

Standard laparoscopy: a single camera, a narrow field of view

Context

During a laparoscopic procedure, the surgeon’s field of view is very narrow: the endoscopic camera, inserted through a trocar, only covers a small portion of the abdominal cavity. The surgeon must constantly reposition the camera to explore the operative field, which is cognitively demanding and lengthens procedures.

The idea: replace a single high-resolution camera with several small, distributed miniature cameras (similar to the phone cameras of the day), and reconstruct a real-time 3D panoramic view of the operative zone.

Standard laparoscopy: a single camera, a narrow field of view

Contributions

This project posed novel challenges at the intersection of computer vision and embedded computing:

Calibration of multiple miniature cameras in an endoscopic configuration
3D reconstruction of the surgical field via distributed stereovision
Fusion of video streams to produce a coherent augmented view
Integration into the CamiTK platform and prototyping for pre-clinical experimentation

3D reconstruction of the operative field from multiple cameras (CamiTK)

The project remained at the research prototype stage but led to a European patent filing.

Skills Applied

Stereoscopic computer vision
Multi-camera optical system calibration
Hardware/software integration
Patent writing

[Patent EP 2016] P. Cinquin, S. Voros, C. Boschet, C. Fouard, A. Moreau-Gaudry — Imaging system for the three dimensional observation of an operation field —

What These Projects Taught Me

These three collaborations were a practical school in medical prototyping — and far more than a checklist of technical skills.

I learned to work with real, imperfect clinical data, often without ground truth available to validate the algorithms. I developed the reflex of designing rigorous evaluation protocols where standard benchmarks didn’t exist, and of integrating operating room constraints from the design phase — time pressure, sterility requirements, ergonomics for the surgeon.

I also shaped my collaborative approach: learning to ask the right questions to clinicians, to translate a medical need into an algorithmic problem, to deliver tools integrated into a software platform usable by non-computer-scientists.

These skills — built alongside a sustained effort in scientific publication and formal knowledge production — are the ones I bring today to medical application prototyping.

Automatic 3D seed location and orientation detection in CT image for prostate brachytherapy

Tue, 29 Apr 2014 00:00:00 +0000

Fully Automatic Organ Localization in Medical Images Using Improved Random Regression Forests

Wed, 01 Jan 2014 00:00:00 +0000

Consideration of seeds orientation in prostate brachytherapy and dosimetric analysis

Sat, 01 Jun 2013 00:00:00 +0000

Non invasive ultrasound-based bone tracking

Wed, 16 Jun 2010 00:00:00 +0000

Weighted distance transforms generalized to modules and their computation on point lattices

Sat, 01 Sep 2007 00:00:00 +0000

An Objective Comparison between Gray Weighted Distance Transforms and Weighted Distance Transforms On Curved Spaces

Wed, 25 Oct 2006 00:00:00 +0000

Generating Distance Maps with Neighbourhood Sequences

Wed, 25 Oct 2006 00:00:00 +0000

Blockwise processing applied to brain microvascular network study

Mon, 25 Sep 2006 00:00:00 +0000

From mathematical foundations to medical images

Tue, 01 Aug 2006 00:00:00 +0000

This project deliberately steps away from applied prototyping. It reflects another side of my approach: the ability to work at a rigorous level of mathematical abstraction, to produce formal proofs, and to publish in leading discrete mathematics journals — while keeping a concrete medical imaging application in view.

From September 2005 to August 2006, I held a postdoctoral position at the at Uppsala University, Sweden — one of the founding laboratories of discrete geometry applied to images. There I had the opportunity to work alongside , a pioneer of the field, whose 1986 paper on distance transforms in digital images remains a landmark reference cited by thousands of researchers — including today in the computation of loss functions for U-Net networks in deep learning segmentation.

The starting point: computing distances in an image

A distance transform is a fundamental operation in image processing: for each pixel of an object, it computes its distance to the nearest boundary. It is ubiquitous — segmentation, skeletonisation, registration, cost function computation in machine learning.

The Euclidean distance is the most natural, but it raises two practical issues:

It is continuous, and therefore expensive to compute exactly on a discrete grid.
The squared Euclidean distance (d²E), which is discrete, does not satisfy the triangle inequality.

This second point deserves an illustration. The triangle inequality simply states that the direct path between two points can never be longer than going via an intermediate stop — it is the fundamental property of any mathematically valid distance.

2D example: d²E(p,q) = 9 > d²E(p,o) + d²E(o,q) = 7. The detour via o is cheaper than the direct path — the shortest path is not the straight line!

The chamfer distance solves this: it is an efficient way to compute distances in a digital image without examining every possible path. Each displacement direction (4-connected, 8-connected neighbour, etc.) is assigned a local weight, and these weights are propagated in a simple two-pass scan of the image. The result is a discrete, integer-valued, fast-to-compute distance — which, under certain conditions on the weights, satisfies all the properties of a mathematical norm.

Chamfer algorithm — 1st pass: propagation left to right, top to bottom

Chamfer algorithm — 2nd pass: propagation right to left, bottom to top

The specific problem: medical images are anisotropic

On an isotropic grid (square pixels, cubic voxels), chamfer distances are well understood. But medical images are rarely isotropic.

Anisotropic medical image: voxels are elongated in the axial direction

A typical CT scan has a resolution of 0.5 mm × 0.5 mm within the axial plane, but 2 to 5 mm between slices. If one naively applies a chamfer algorithm designed for an isotropic grid to such an image, the computed distances no longer satisfy the triangle inequality — and can lead to geometric absurdities where the direct path between two points appears longer than a detour.

The standard workaround is to resample the image to make it isotropic before computing distances. But this resampling introduces artefacts and significantly increases processing overhead.

My contribution: a unified theory for arbitrary grids

My postdoctoral work consisted in developing a general mathematical framework for computing chamfer distances directly on any grid — in particular the anisotropic grids of medical images — with the formal guarantee that the resulting distance is a genuine norm in the mathematical sense (positive, symmetric, triangle inequality, positive homogeneity).

The key insight is the notion of a module: by generalising the standard discrete grid framework (Z²) to that of modules over commutative rings, I was able to establish the necessary and sufficient conditions on the chamfer mask weights to guarantee the norm property — independently of the grid geometry.

This work led to three major contributions:

1. Distances on general grids — proof of the norm

The main paper in Pattern Recognition (2007) establishes the complete theory: definitions, properties (distance, metric, norm), validity conditions for the sequential two-scan algorithm, and application to FCC (face-centered cubic) and BCC (body-centered cubic) grids — crystallographic structures with optimal sampling properties for 3D medical imaging.

My contribution: I was the primary author of this paper. I developed the generalisation to modules, established the validity conditions for the two-pass algorithm on arbitrary grids, and computed the optimal weights for FCC and BCC grids.

[Pattern Recognition 2007] C. Fouard, R. Strand, G. Borgefors — Weighted distance transforms generalized to modules and their computation on point lattices —

2. Distances on non-standard grids via neighbourhood sequences

An extension to distances defined by neighbourhood sequences (which allow even better isotropy), with formal proof of the conditions for the sequential algorithm to produce correct distance maps on square, cubic, FCC, and BCC grids.

[DGCI 2006] R. Strand, B. Nagy, C. Fouard, G. Borgefors — Generating distance maps with neighbourhood sequences —

3. Comparison of grey-level distance transforms

A comparative study of two distance definitions on grey-level images (GWDT and WDTOCS), with theoretical and experimental analysis of their respective behaviours on different image types (density maps, height maps).

[DGCI 2006] C. Fouard, M. Gedda — Distance transforms on curved spaces —

What this work gave me

Those twelve months in Uppsala taught me to hold two stances simultaneously: that of the mathematician who proves, and that of the engineer who solves a concrete problem. Formal rigour is not an end in itself — it is what guarantees that an algorithm will behave correctly on real data, including in the edge cases one did not anticipate.

The indirect impact of this work on medical imaging practice is real: most modern segmentation tools, including U-Net networks that compute their loss functions from distance maps, build on these theoretical foundations. I have not yet had the opportunity to apply these anisotropic distances directly in my applied research projects in Grenoble — but it is a direction I keep in mind, particularly for processing medical images without prior resampling.

A Novel Three-Dimensional Computer-Assisted Method for a Quantitative Study of Microvascular Networks of the Human Cerebral Cortex

Sun, 01 Jan 2006 00:00:00 +0000

3-D chamfer distances and norms in anisotropic grids

Tue, 01 Feb 2005 00:00:00 +0000

Extraction de paramètres morphométriques pour l'étude du réseau micro-vasculaire cérébral

Fri, 21 Jan 2005 00:00:00 +0000

Extraction of morphometric parameters for the study of the cerebral micro-vascular network

Fri, 21 Jan 2005 00:00:00 +0000

Measuring the brain in 3D: from microscopy to software tools.

My PhD took place at INRIA Sophia Antipolis, in the Epidaure team, under the supervision of Grégoire Malandain. It was part of the MicroVisu3D project, which brought together three worlds: the anatomists at INSERM (unit U455), who needed to quantify cerebral microcirculation; the image-analysis researchers at INRIA; and an industrial partner, TGS Europe, publisher of the 3D visualization software Amira.

It was a CIFRE thesis, funded by the company: from the very start of my doctorate, I worked at the direct interface between academic research and industry. My role was to translate a scientific need — “to be able to measure the cortical vascular network in 3D” — into a concrete set of software tools, usable by non-programmers and integrable into an existing industrial environment.

This work led to publications in international peer-reviewed journals and conferences. But beyond the scientific output, what I describe here is an experience of applied R&D delivered end to end — from gathering the need to shipping field-tested tools.

This thesis embodies an idea that still guides my work today: an algorithm is only worth as much as its robustness on real data — imperfect, voluminous, and all different from one another.

The challenge: measuring a vast network at micron scale

To study the cerebral micro-vascular network, the anatomists needed images able to capture the smallest capillary (resolution of a few microns) over a cortical surface large enough to be statistically significant (on the order of a centimetre). No instrument could acquire such an image in a single shot.

The chosen solution: tile the area to be imaged with many small images acquired under a confocal microscope, then assemble them into a single large “image mosaic”. This solved the acquisition problem, but created two new ones:

the resulting mosaic was too large to be loaded into the memory of a standard computer and processed in one go;
the confocal microscope imposes an anisotropic grid (voxels are not cubic), which complicates any distance measurement in the image — and therefore any computation of vessel diameter.

A mosaic of 118 confocal microscopy images covering a cortical sulcus (≈ 0.8 × 0.8 cm)

My approach: an end-to-end processing pipeline

I designed a complete pipeline, from the sensor to the measurement, built from the outset to run on data that does not fit in memory.

flowchart LR A["Acquisition
confocal mosaic"] --> B["Registration &
realignment"] B --> C["Single
virtual image"] C --> D["Block-wise processing
(out-of-core)"] D --> E["Distance maps
+ centerlines"] E --> F["Measurement &
3D visualization"]

Two links in this pipeline gave rise to original methodological contributions: the computation of distance maps and the extraction of centerlines.

Contribution 1 — Distance maps that adapt on their own

To measure the radius of a vessel at each point, one computes a distance map: at each voxel, the distance to the nearest boundary. The exact Euclidean distance is costly; chamfer distances approximate it very efficiently by propagating small integer weights between neighbouring voxels.

A binary shape and its distance map: at each point, the distance to the nearest boundary

The tricky part: on an anisotropic grid — the case for almost every medical imaging modality — these weights have to be recomputed, and choosing them by hand is fragile and modality-specific.

My contribution was to propose an automatic computation of the optimal chamfer coefficients, the set that minimizes the error against the true Euclidean distance — and this whatever the grid anisotropy or the imaging modality, with no manual tuning. The same method therefore applies equally to a CT scan, an MRI or a confocal microscope.

The theoretical foundations of these chamfer masks were deepened during my post-doctorate: see .

Contribution 2 — Extracting centerlines, even out-of-core

From the distance map, one extracts the centerlines of the vessels — the curves running through the centre of each vessel. They capture the topology of the network and make it possible to measure lengths, branchings and radii.

Classical skeletonization methods require loading the whole image into memory — impossible here. So I proposed a block-wise skeletonization that works on sub-images while preserving the three essential properties of a skeleton: homotopy (same topology as the original object), localization (the skeleton stays centred) and thinness. The algorithm also minimizes the number of sub-image accesses, to keep computation time acceptable.

Skeletonization of an object: without tiling (a), by processing blocks separately (b), and with our method (c), which preserves topology

Once centerlines and radii are obtained, the vessels are modelled as sets of cylinders (the LineSet data structure), allowing both real-time 3D visualization and the extraction of quantitative parameters: distributions of diameters, of lengths, vascular densities per cortical layer, and so on.

Overview of the centerlines of the reconstructed micro-vascular network in 3D

A generic method, far beyond the brain

None of these methods makes any assumption specific to the brain. Any large binary image of tubular structures can be processed in the same way. I validated it on plant roots; it transfers directly to neural networks, to the porosity of materials, even to pipeline mapping.

The same method applied to plant roots: concrete proof of the generality of the approach

This ability to solve a problem in one domain and then transpose it elsewhere is at the heart of how I approach prototyping.

Skills involved

Gathering a scientific need and translating it into software specifications
Collaboration with an industrial partner (CIFRE thesis, integration into the Amira software / TGS Europe)
Design of robust, automatic algorithms, with no manual tuning
Processing large, out-of-core data (block by block)
Discrete geometry, chamfer distances, discrete topology
Design of validation protocols (synthetic + real data)

Publications from the thesis

2007. Fouard Céline, Strand Robin, Borgefors Gunilla — Weighted distance transforms generalized to modules and their computation on point lattices. Pattern Recognition Vol 40 Issue 9
2006. Fouard Céline, Gedda Magnus — An Objective Comparison between Gray Weighted Distance Transforms and Weighted Distance Transforms On Curved Spaces. Proceedings of DGCI 2006
2006. Fouard Céline, Malandain Grégoire, Prohaska Steffen, Westerhoff Malte — Blockwise processing applied to brain microvascular network study. *IEEE Transactions on Medical Imaging ( Volume: 25, Issue: 10, October 2006) *
2006. Cassot Francis, Lauwers Frederic, Fouard celineProhaska Steffen, Lauwers-Cance Valérie — A Novel Three-Dimensional Computer-Assisted Method for a Quantitative Study of Microvascular Networks of the Human Cerebral Cortex. Microcirculation
2005. Fouard Céline, Malandain Grégoire — 3-D chamfer distances and norms in anisotropic grids. Image and Vision Computing
2005. Fouard Céline — Extraction de paramètres morphométriques pour l'étude du réseau micro-vasculaire cérébral. Thèse de doctorat, Université de Nice Sophia Antipolis
2004. Kolesik Peter, Fouard Céline, Prohaska Steffen, McNeill Ann — Automated method for non-destructive 3D visualisation of plant root architecture using X-ray tomography. 4th International Workshop on Functional-Structural Plant Models
2003. Fouard Céline, Malandain Grégoire — Systematized calculation of optimal coefficients of 3-D chamfer norms. Discrete Geometry for Computer Imagery, 11th International Conference, DGCI 2003

What this thesis taught me

This thesis was, for me, a first school of applied prototyping — far more than a theoretical exercise.

Because it was a CIFRE thesis, I first learned to work between several worlds: to ask anatomists the right questions, to translate their need into an algorithmic problem, and to deliver a result that could be integrated into an industrial partner’s software. This stance — a translator between use and technique — has stayed at the centre of my practice.

I then built the reflex of designing robust, automatic methods. Faced with data that was all different — varying anisotropic grids, uneven contrast — I sought algorithms that adapt by themselves rather than multiplying manual settings, a source of fragility and variability.

I also developed a taste for generality: building tools that outgrow their initial use case. The same building blocks, designed for the brain, proved useful for plant roots — and many other domains.

Finally, I learned to deal with a very concrete constraint: data too large for memory. Thinking of computation piece by piece, without ever sacrificing the correctness of the result, is a skill that serves in any project handling large volumes.

These skills — built alongside a sustained effort of scientific publication — are the ones I now put at the service of medical-application prototyping.

Systematized calculation of optimal coefficients of 3-D chamfer norms

Sat, 01 Nov 2003 00:00:00 +0000

Interventional cardiology

Mon, 01 Jan 0001 00:00:00 +0000

From clinical need to prototype: guiding the gesture at the heart of the cath lab.

Interventional cardiology is performed under imaging, yet the clinician must often act without directly seeing the target: the lesion to treat or to biopsy does not appear on the real-time image in the room. Carried out with Prof. Gilles Barone-Rochette (Grenoble Alpes University Hospital) and the LTSI laboratory in Rennes, this project pursues a single goal expressed through two clinical questions: to give the cardiologist reliable guidance, built from preoperative imaging and usable directly in the room.

The common thread: bouncing back when the data runs short

The project started with the guidance of cell therapy. The first clinical trial enrolled fewer patients than expected: the data needed for the next steps were not there. Rather than abandoning it, we redeployed the technical building blocks already developed (image segmentation, navigation) toward a related clinical need with more immediate value and a better-identified bottleneck: endomyocardial biopsy.

This pivot is, in itself, a deliverable: it illustrates the ability to de-risk a project, preserve the assets already built, and refocus the effort where the clinical value is highest — exactly the kind of trade-off a company expects when an R&D programme does not go as planned.

Sub-project 1 — Guiding post-infarction cell therapy

After a heart attack, some therapies consist in re-injecting cells into the myocardium. The whole challenge is precision: reaching the right areas, relying on information (the extent of fibrosis, the viable regions) that is only visible on preoperative imaging, not on the real-time image in the room.

Our approach: fuse multimodal imaging to transfer, during the intervention, the targets identified preoperatively. The core building block is the automatic segmentation of the myocardium and fibrosis on late gadolinium enhancement MRI (LGE-MRI), developed using deep learning as part of Erwan Lecesne’s PhD (co-supervised with the LTSI in Rennes), then integrated into to be presented to the clinician in the room.

Re-injecting cells in the right place: the precision of the gesture determines the therapy’s effectiveness.

CamiTK is a prototyping toolkit: it makes it possible to move quickly from concept to evaluated prototype, but its output is not meant to be a CE-marked medical device. This building block is therefore a proof of concept; its industrial transfer is currently under discussion with partners in the field.

Sub-project 2 — A map for endomyocardial biopsy

Three conditions — cardiac sarcoidosis, chronic myocarditis and arrhythmogenic cardiomyopathy — can present a similar clinical picture yet call for opposite treatments. To decide, a biopsy is needed… provided the sample is taken in the right place.

Endomyocardial biopsy: sampling a piece of cardiac tissue, where the lesion is located.

The state of the art leaves a real gap:

“blind” biopsy is poorly specific, because the fibrosis to target stays invisible during the gesture;
electro-anatomical guidance is long and, likewise, blind to fibrosis.

Our solution acts as a “GPS” for the biopsy catheter: it transfers the target identified on preoperative imaging onto the real-time image, to guide the sampling. Two design choices make it a solution built for adoption: it is hardware-independent (compatible with an existing room) and works on real-time fluoroscopy, with no complex fusion step. It directly reuses the segmentation building block from the first sub-project.

Diagram of the proposed guidance system for endomyocardial biopsy (published).

Current status: we are preparing the first clinical trials in the laboratory; industrial transfer will follow.

What this project demonstrates

Designing from a real clinical need, in close dialogue with practitioners, rather than around a technical feat.
Mastering medical image processing and AI, and putting them at the service of a precise, useful target.
Taking a prototype from the laboratory toward the clinic, with a clear awareness of maturity stages (TRL), the trial framework and CE marking.
Architecting for adoption: hardware independence, integration into a prototyping toolkit, reuse of building blocks.
Refocusing a project to preserve its value: agility and de-risking in the face of the unexpected.
Leading a multi-site collaboration (Grenoble–Rennes) and co-supervising a PhD.

Funding obtained

Funding	Amount	Purpose
Famtastic project (France Life Imaging)	€20,000	Kick-starting the collaboration with the LTSI (Rennes)
PUI (UGA)	€60,000	Maturing the prototype toward the first clinical trials
PhD funding (LabeX CAMI)	€170,000	Co-supervision of Erwan Lecesne’s PhD
Post-doctorate (LabeX CAMI)	€56,000	One year of post-doctoral engineering
Total	€306,000

Collaborations and supervision

Prof. Gilles Barone-Rochette — interventional cardiologist, Grenoble Alpes University Hospital: clinical partner of the project.
LTSI laboratory (Rennes) — Professor Mireille Garreau and Antoine Simon (associate professor): collaboration on cardiac image processing.
Erwan Lecesne’s PhD (2020–2024), co-supervised at 50% with Mireille Garreau (LTSI): multimodal image processing to improve post-infarction cell therapy.
Théophile Tiffet’s PhD — medical resident: echocardiography / SPECT calibration for interventional cardiology.

2024. Barone-Rochette Gilles, MD,, Erwan Lecesne, MSc,, Antoine Simon, PhD, Mireille Garreau, PhD,, Celine Fouard, PhD — New Method CMR-Guided Endomyocardial Biopsy in Suspicion Context of Isolated Cardiac Sarcoidosis. Circulation: Cardiovascular Imaging, vol 17, no 4
2023. Erwan Lecesne, Antoine Simon, Mireille Garreau, Barone-Rochette Gilles, Celine Fouard — Segmentation of cardiac infarction in delayed-enhancement MRI using probability map and transformers-based neural networks. Computer Methods and Programs in Biomedicine, vol 242
2023. Lecesne Erwan, Simon Antoine, Garreau Mireille, Barone-Rochette Gilles, Fouard Céline — Transformers-Based Neural Network for Cardiac Infarction Segmentation in Delayed-Enhancement MRI. 2023 IEEE Twelfth International Conference on Image Processing Theory, Tools and Applications (IPTA)

Analyse-Images-Medicales |

New Method CMR-Guided Endomyocardial Biopsy in Suspicion Context of Isolated Cardiac Sarcoidosis

Segmentation of cardiac infarction in delayed-enhancement MRI using probability map and transformers-based neural networks

Transformers-Based Neural Network for Cardiac Infarction Segmentation in Delayed-Enhancement MRI

Light Random Regression Forests for automatic multi-organ localization in CT images

Automatically locating organs in CT images

The challenge: locating, not segmenting

The contribution

Robustness — and bias: the phantom-kidney story

What this project represented

Related publications

Segmentation, separation and pose estimation of prostate brachytherapy seeds in CT images

Impact de l’orientation des grains iode 125 dans l’évaluation de la dosimétrie à 1 mois d’une curiethérapie prostatique

Medical Image Analysis

Prostate Brachytherapy — Orientation of Iodine-125 Seeds

Context

Contributions

Skills Applied

Related Publications

3D Ultrasound-Based Bone Tracking

Context

Contributions

Skills Applied

Related Publications

Distributed Laparoscopy — Extended View of the Surgical Field

Context

Contributions

Skills Applied

Related Output

What These Projects Taught Me

Automatic 3D seed location and orientation detection in CT image for prostate brachytherapy

Fully Automatic Organ Localization in Medical Images Using Improved Random Regression Forests

Consideration of seeds orientation in prostate brachytherapy and dosimetric analysis

Non invasive ultrasound-based bone tracking

Weighted distance transforms generalized to modules and their computation on point lattices

An Objective Comparison between Gray Weighted Distance Transforms and Weighted Distance Transforms On Curved Spaces

Generating Distance Maps with Neighbourhood Sequences

Blockwise processing applied to brain microvascular network study

From mathematical foundations to medical images

The starting point: computing distances in an image

The specific problem: medical images are anisotropic

My contribution: a unified theory for arbitrary grids

1. Distances on general grids — proof of the norm

2. Distances on non-standard grids via neighbourhood sequences

3. Comparison of grey-level distance transforms

What this work gave me

A Novel Three-Dimensional Computer-Assisted Method for a Quantitative Study of Microvascular Networks of the Human Cerebral Cortex

3-D chamfer distances and norms in anisotropic grids

Extraction de paramètres morphométriques pour l'étude du réseau micro-vasculaire cérébral

Extraction of morphometric parameters for the study of the cerebral micro-vascular network

The challenge: measuring a vast network at micron scale

My approach: an end-to-end processing pipeline

Contribution 1 — Distance maps that adapt on their own

Contribution 2 — Extracting centerlines, even out-of-core

A generic method, far beyond the brain

Skills involved

Publications from the thesis

What this thesis taught me

Systematized calculation of optimal coefficients of 3-D chamfer norms

Interventional cardiology

The common thread: bouncing back when the data runs short

Sub-project 1 — Guiding post-infarction cell therapy

Sub-project 2 — A map for endomyocardial biopsy

What this project demonstrates

Funding obtained

Collaborations and supervision

Related publications