Les commentaires sont clos.

[INSERM/DSI/CISI] Numerical reliability of computation codes in medical imaging

12 Avril 2016

Catégorie : Journée d étude

La coordination de l'informatique scientifique de l'INSERM (DSI/CISI) propose le mercredi 27 avril un séminaire sur le calcul en imagerie intitulé « Numerical reliability of computation codes in medical imaging ». Ce séminaire aura lieu à l’auditorium du Biopark (11 rue Watt) à Paris de 13h30 à 18h.

Vous trouverez ci-dessous un résumé des présentations prévues.

Les inscriptions sont gratuites et obligatoires ici :


Title: Numerical reliability of computation codes in medical imaging

In the context of high performance computing, new architectures, becoming more and more parallel, offer higher floating-point computing power. Efficient simulation programs that take benefit of this computing power are developed for medical imaging. Because of the higher level of parallelism and of the large number of floating-point operations, numerical validation of these programs becomes increasingly important. The aim of this seminar is a discussion about the numerical reliability of high performance simulations for medical imaging. New high performance algorithms for medical imaging will be presented, as topics related to numerical validation on high performance architectures such as numerical reproducibility or precision optimisation.

1- Fast and accurate medical image simulation using GPU-accelerated computing

Julien Bert, CHRU de Brest, LaTIM, INSERM UMR1101

Monte Carlo simulations (MCS) are using random sampling methods for representing and solving physical and mathematical problems. They play a key role in medical applications, both for imaging and radiotherapy by accurately modeling the different physical processes and interactions between particles and matter (tissues and/or detectors). In the medical imaging field, MCS are used in the design of imaging systems, optimization of acquisition protocols, as well as in the development and assessment of image reconstruction processes and associated correction algorithms. However, MCS are also associated with long execution times, which is one of the major issues preventing their use in research and clinical environment. Recently, graphics processing units (GPU) have become in many different domains a low cost alternative solution for the acquisition of high computation power. However, the computational power of such system may be at the cost of the numerical reliability. New algorithms and strategies have to be developed in order to reach the accuracy required for medical applications. A brief introduction on MCS will be presented showing why such approach needs numerical accuracy. An introduction on GPU architecture will follow and examples of implementation will be detailed through two applications, one in Positron Emission Tomography and one in X-ray imaging. Finally, solutions that improve the numerical reliability on GPU architecture will be presented and discussed.

2- Estimation of numerical reproducibility using stochastic arithmetic (Fabienne Jézéquel, LIP6)

Fabienne Jézéquel, LIP6 (Laboratoire d'Informatique de Paris 6)

Differences in simulation results may be observed from one architecture to another or even inside the same architecture. Such reproducibility failures are often due to different rounding errors generated by different orders in the sequence of arithmetic operations. It must be pointed out that the cause of differences in results may be difficult to identify: rounding errors or bug? Such differences are particularly noticeable with multicore processors or
GPUs (Graphics Processing Units). In this talk, we describe the principles of DSA (Discrete Stochastic Arithmetic) which enables one to estimate rounding error propagation in simulation programs. We show that DSA can be used to estimate which digits in simulation results may be different from one environment to another because of rounding errors. We present the CADNA library (, an implementation of DSA that controls the numerical quality of programs and detects numerical instabilities generated during the execution. A particular version of CADNA which enables numerical validation in hybrid CPU-GPU environments is described. The estimation of numerical reproducibility using DSA is illustrated by a wave propagation code which can be affected by reproducibility problems when executed on different architectures.

3- Towards scalable methods in biomedical imaging and digital pathology

Daniel Racoceanu, Sorbonne Universités, UPMC Univ Paris 06, CNRS, INSERM, Laboratoire d'Imagerie Biomédicale (LIB)
Co-speakers : Sreetama Basu, PhD (Institut de Biologie, Ecole Normale Supérieure - IBENS), Daniel Salas (INSERM/DSI/CISI), Soufianne Haddani (Univ. Paris Decartes et LIB - UMR CNRS / U INSERM)

Being able to bring justified and traceable responses has become an ethical priority for the patients and the healthcare professionals in modern medicine.

After initiating – during the last few years, - a semantic cognitive virtual microscopy approach for routine-pathology-compliant protocols dedicated to mitotic count and nuclear atypia and some pioneer international benchmarking initiatives (Int. Conf. Pattern Recognition ICPR 2012 : MITOS – Tsukuba, Japan and ICPR 2014 : MITOS & ATYPIA – Stockholm, Sweden), our present priority is to design an integrative microscopy framework able to combine reasoning, medical ontologies, knowledge-based WSI (Whole Slide Images) exploration rules with scalable imaging protocols.

Therefore, our research focus on two families of methods based on a strong mathematical formalism: one (radiometric) inspired by Marked Point Process and one (structural) based on Point-set Mathematical Morphology. In addition to be scalable and efficient, these methods are very promising with regard to integrative approaches as the big data challenge, both inherent to the future of Digital Pathology and, in general, to biomedical imaging.

This talk will focus on the Marked Point Process and the efforts we did deploy to adapt this method to neurite tracing in 3D confocal imaging and to nuclear atypia assessment in breast cancer histopathology. Challenges related to parameter estimation, reconstruction, parallelisation and validation will be highlighted during the discussions.