Data infrastructures for AI in medical imaging: a report on the experiences of five EU projects.

Artificial intelligence (AI) is transforming the field of medical imaging and has the potential to bring medicine from the era of 'sick-care' to the era of healthcare and prevention. The development of AI requires access to large, complete, and harmonized real-world datasets, representative of the population, and disease diversity. However, to date, efforts are fragmented, based on single-institution, size-limited, and annotation-limited datasets. Available public datasets (e.g., The Cancer Imaging Archive, TCIA, USA) are limited in scope, making model generalizability really difficult. In this direction, five European Union projects are currently working on the development of big data infrastructures that will enable European, ethically and General Data Protection Regulation-compliant, quality-controlled, cancer-related, medical imaging platforms, in which both large-scale data and AI algorithms will coexist. The vision is to create sustainable AI cloud-based platforms for the development, implementation, verification, and validation of trustable, usable, and reliable AI models for addressing specific unmet needs regarding cancer care provision. In this paper, we present an overview of the development efforts highlighting challenges and approaches selected providing valuable feedback to future attempts in the area.Key points• Artificial intelligence models for health imaging require access to large amounts of harmonized imaging data and metadata.• Main infrastructures adopted either collect centrally anonymized data or enable access to pseudonymized distributed data.• Developing a common data model for storing all relevant information is a challenge.• Trust of data providers in data sharing initiatives is essential.• An online European Union meta-tool-repository is a necessity minimizing effort duplication for the various projects in the area.

Pharmacological Characterization of [18F]-FNM and Evaluation of NMDA Receptors Activation in a Rat Brain Injury Model.

PURPOSE: NMDA receptors (NMDARs) dysfunction plays a central role in the physiopathology of psychiatric and neurodegenerative disorders whose mechanisms are still poorly understood. The development of a PET (positron emission tomography) tracer able to selectively bind to the NMDARs intra-channel PCP site may make it possible to visualize NMDARs in an open and active state. We describe the in vitro pharmacological characterization of [18F]-fluoroethylnormemantine ([18F]-FNM) and evaluate its ability to localize activated NMDA receptors in a rat preclinical model of excitotoxicity. PROCEDURES: The affinity of the non-radioactive analog for the intra-channel PCP site was determined in a radioligand competition assay using [3H]TCP ([3H]N-(1-[thienyl]cyclohexyl)piperidine) on rat brain homogenates. Selectivity was also investigated by the displacement of specific radioligands targeting various cerebral receptors. In vivo brain lesions were performed using stereotaxic quinolinic acid (QA) injections in the left motor area (M1) of seven Sprague Dawley rats. Each rat was imaged with a microPET/CT camera, 40 min after receiving a dose of 30 MBq + / - 20 of [18F]-FNM, 24 and 72 h after injury. Nine non-injured rats were also imaged using the same protocol. RESULTS: FNM displayed IC50 value of 13.0 ± 8.9 µM in rat forebrain homogenates but also showed significant bindings on opioid receptors. In the frontal and left somatosensory areas, [18F]FNM PET detected a mean of 37% and 41% increase in [18F]FNM uptake (p < 0,0001) 24 and 72 h after QA stereotaxic injection, respectively, compared to the control group. CONCLUSIONS: In spite of FNM's poor affinity for NMDAR PCP site, this study supports the ability of this tracer to track massive activation of NMDARs in neurological diseases.

Ex vivo susceptibility-weighted imaging anatomy of canine brain-comparison of imaging and histological sections.

Now that access of large domestic mammals to high-field MRI becomes more common, techniques initially implemented for human patients can be used for the structural and functional study of the brain of these animals. Among them, susceptibility-weighted imaging (SWI) is a recent technique obtained from gradient echo (GE) imaging that allow for an excellent anatomical tissue contrast and a non-invasive assessment of brain iron content. The goal of this study was to design an optimal GE SWI imaging protocol to be used in dogs undergoing an MRI examination of the brain in a 3-Tesla scanner. This imaging protocol was applied to ex vivo brains from four dogs. The imaging protocol was validated by visual inspection of the SWI images that provided a high anatomical detail, as demonstrated by their comparison with corresponding microscopic sections. As resolvable brain structures were labeled, this study is the first to provide an anatomic description of SWI images of the canine brain. Once validated in living animals, this GE SWI imaging protocol could be easily included in routine neuroimaging protocols to improve the diagnosis of various intracranial diseases of dogs, or be used in future comparative studies aiming at evaluating brain iron content in animals.

A loss-of-function mutation in RORB disrupts saltatorial locomotion in rabbits.

Saltatorial locomotion is a type of hopping gait that in mammals can be found in rabbits, hares, kangaroos, and some species of rodents. The molecular mechanisms that control and fine-tune the formation of this type of gait are unknown. Here, we take advantage of one strain of domesticated rabbits, the sauteur d'Alfort, that exhibits an abnormal locomotion behavior defined by the loss of the typical jumping that characterizes wild-type rabbits. Strikingly, individuals from this strain frequently adopt a bipedal gait using their front legs. Using a combination of experimental crosses and whole genome sequencing, we show that a single locus containing the RAR related orphan receptor B gene (RORB) explains the atypical gait of these rabbits. We found that a splice-site mutation in an evolutionary conserved site of RORB results in several aberrant transcript isoforms incorporating intronic sequence. This mutation leads to a drastic reduction of RORB-positive neurons in the spinal cord, as well as defects in differentiation of populations of spinal cord interneurons. Our results show that RORB function is required for the performance of saltatorial locomotion in rabbits.

Diffusion Tensor Imaging Tractography of White Matter Tracts in the Equine Brain.

Tractography, a noninvasive technique tracing brain pathways from diffusion tensor magnetic resonance imaging (DTI) data, is increasingly being used for brain investigation of domestic mammals. In the equine species, such a technique could be useful to improve our knowledge about structural connectivity or to assess structural changes of white matter tracts potentially associated with neurodegenerative diseases. The goals of the present study were to establish the feasibility of DTI tractography in the equine brain and to provide a morphologic description of the most representative tracts in this species. Postmortem DTI and susceptibility-weighted imaging (SWI) of an equine brain were acquired with a 3-T system using a head coil. Association, commissural, and projection fibers, the three fiber groups typically investigated in tractography studies, were successfully reconstructed and overlaid on SWI or fractional anisotropy maps. The fibers derived from DTI correlate well with their description in anatomical textbooks. Our results demonstrate the feasibility of using postmortem DTI data to reconstruct the main white matter tracts of the equine brain. Further DTI acquisitions and corresponding dissections of equine brains will be necessary to validate these findings and create an equine stereotaxic white matter atlas that could be used in future neuroimaging research.

In vivo localization of cortical areas using a 3D computerized atlas of the marmoset brain.

We created a volumetric template of the marmoset (Callithrix jacchus) brain, which enables localization of the cortical areas defined in the Paxinos et al. (The marmoset brain in stereotaxic coordinates. Elsevier Academic Press, Cambridge, 2012) marmoset brain atlas, as well as seven broader cortical regions (occipital, temporal, parietal, prefrontal, motor, limbic, insular), different brain compartments (white matter, gray matter, cerebro-spinal fluid including ventricular spaces), and various other structures (brain stem, cerebellum, olfactory bulb, hippocampus). The template was designed from T1-weighted MR images acquired using a 3 T MRI scanner. It was based on a single fully segmented marmoset brain image, which was transported onto the mean of 13 adult marmoset brain images using a diffeomorphic strategy that fully preserves the brain topology. In addition, we offer an automatic segmentation pipeline which fully exploits the proposed template. The segmentation pipeline was quantitatively assessed by comparing the results of manual and automated segmentations. An associated program, written in Python, can be used from a command-line interface, or used interactively as a module of the 3DSlicer software. This program can be applied to the analysis of multimodal images, to map specific cortical areas in lesions or to define the seeds for further tractography analyses.

Dwarfism and Altered Craniofacial Development in Rabbits Is Caused by a 12.1 kb Deletion at the HMGA2 Locus.

The dwarf phenotype characterizes the smallest of rabbit breeds and is governed largely by the effects of a single dwarfing allele with an incompletely dominant effect on growth. Dwarf rabbits typically weigh under 1 kg and have altered craniofacial morphology. The dwarf allele is recessive lethal and dwarf homozygotes die within a few days of birth. The dwarf phenotype is expressed in heterozygous individuals and rabbits from dwarf breeds homozygous for the wild-type allele are normal, although smaller when compared to other breeds. Here, we show that the dwarf allele constitutes a ∼12.1 kb deletion overlapping the promoter region and first three exons of the HMGA2 gene leading to inactivation of this gene. HMGA2 has been frequently associated with variation in body size across species. Homozygotes for null alleles are viable in mice but not in rabbits and probably not in humans. RNA-sequencing analysis of rabbit embryos showed that very few genes (4-29 genes) were differentially expressed among the three HMGA2/dwarf genotypes, suggesting that dwarfism and inviability in rabbits are caused by modest changes in gene expression. Our results show that HMGA2 is critical for normal expression of IGF2BP2, which encodes an RNA-binding protein. Finally, we report a catalog of regions of elevated genetic differentiation between dwarf and normal-size rabbits, including LCORL-NCAPG, STC2, HOXD cluster, and IGF2BP2 Levels and patterns of genetic diversity at the LCORL-NCAPG locus further suggest that small size in dwarf breeds was enhanced by crosses with wild rabbits. Overall, our results imply that small size in dwarf rabbits results from a large effect, loss-of-function (LOF) mutation in HMGA2 combined with polygenic selection.

Levels and Patterns of Genetic Diversity and Population Structure in Domestic Rabbits.

Over thousands of years humans changed the genetic and phenotypic composition of several organisms and in the process transformed wild species into domesticated forms. From this close association, domestic animals emerged as important models in biomedical and fundamental research, in addition to their intrinsic economical and cultural value. The domestic rabbit is no exception but few studies have investigated the impact of domestication on its genetic variability. In order to study patterns of genetic structure in domestic rabbits and to quantify the genetic diversity lost with the domestication process, we genotyped 45 microsatellites for 471 individuals belonging to 16 breeds and 13 wild localities. We found that both the initial domestication and the subsequent process of breed formation, when averaged across breeds, culminated in losses of ~20% of genetic diversity present in the ancestral wild population and domestic rabbits as a whole, respectively. Despite the short time elapsed since breed diversification we uncovered a well-defined structure in domestic rabbits where the FST between breeds was 22%. However, we failed to detect deeper levels of structure, probably consequence of a recent and single geographic origin of domestication together with a non-bifurcating process of breed formation, which were often derived from crosses between two or more breeds. Finally, we found evidence for intrabreed stratification that is associated with demographic and selective causes such as formation of strains, colour morphs within the same breed, or country/breeder of origin. These additional layers of population structure within breeds should be taken into account in future mapping studies.

Rabbit genome analysis reveals a polygenic basis for phenotypic change during domestication.

The genetic changes underlying the initial steps of animal domestication are still poorly understood. We generated a high-quality reference genome for the rabbit and compared it to resequencing data from populations of wild and domestic rabbits. We identified more than 100 selective sweeps specific to domestic rabbits but only a relatively small number of fixed (or nearly fixed) single-nucleotide polymorphisms (SNPs) for derived alleles. SNPs with marked allele frequency differences between wild and domestic rabbits were enriched for conserved noncoding sites. Enrichment analyses suggest that genes affecting brain and neuronal development have often been targeted during domestication. We propose that because of a truly complex genetic background, tame behavior in rabbits and other domestic animals evolved by shifts in allele frequencies at many loci, rather than by critical changes at only a few domestication loci.

Emergence of a new lagovirus related to Rabbit Haemorrhagic Disease Virus.

Since summer 2010, numerous cases of Rabbit Haemorrhagic Disease (RHD) have been reported in north-western France both in rabbitries, affecting RHD-vaccinated rabbits, and in wild populations. We demonstrate that the aetiological agent was a lagovirus phylogenetically distinct from other lagoviruses and which presents a unique antigenic profile. Experimental results show that the disease differs from RHD in terms of disease duration, mortality rates, higher occurrence of subacute/chronic forms and that partial cross-protection occurs between RHDV and the new RHDV variant, designated RHDV2. These data support the hypothesis that RHDV2 is a new member of the Lagovirus genus. A molecular epidemiology study detected RHDV2 in France a few months before the first recorded cases and revealed that one year after its discovery it had spread throughout the country and had almost replaced RHDV strains. RHDV2 was detected in continental Italy in June 2011, then four months later in Sardinia.

Hepatitis E virus strains in rabbits and evidence of a closely related strain in humans, France.

Hepatitis E virus (HEV) strains from rabbits indicate that these mammals may be a reservoir for HEVs that cause infection in humans. To determine HEV prevalence in rabbits and the strains' genetic characteristics, we tested bile, liver, and additional samples from farmed and wild rabbits in France. We detected HEV RNA in 7% (14/200) of bile samples from farmed rabbits (in 2009) and in 23% (47/205) of liver samples from wild rabbits (in 2007-2010). Full-length genomic sequences indicated that all rabbit strains belonged to the same clade (nucleotide sequences 72.2%-78.2% identical to HEV genotypes 1-4). Comparison with HEV sequences of human strains and reference sequences identified a human strain closely related to rabbit strain HEV. We found a 93-nt insertion in the X domain of open reading frame 1 of the human strain and all rabbit HEV strains. These findings indicate that the host range of HEV in Europe is expanding and that zoonotic transmission of HEV from rabbits is possible.

The genetic structure of domestic rabbits.

Understanding the genetic structure of domestic species provides a window into the process of domestication and motivates the design of studies aimed at making links between genotype and phenotype. Rabbits exhibit exceptional phenotypic diversity, are of great commercial value, and serve as important animal models in biomedical research. Here, we provide the first comprehensive survey of nucleotide polymorphism and linkage disequilibrium (LD) within and among rabbit breeds. We resequenced 16 genomic regions in population samples of both wild and domestic rabbits and additional 35 fragments in 150 rabbits representing six commonly used breeds. Patterns of genetic variation suggest a single origin of domestication in wild populations from France, supporting historical records that place rabbit domestication in French monasteries. Levels of nucleotide diversity both within and among breeds were ~0.2%, but only 60% of the diversity present in wild populations from France was captured by domestic rabbits. Despite the recent origin of most breeds, levels of population differentiation were high (F(ST) = 17.9%), but the majority of polymorphisms were shared and thus transferable among breeds. Coalescent simulations suggest that domestication began with a small founding population of less than 1,200 individuals. Taking into account the complex demographic history of domestication with two successive bottlenecks, two loci showed deviations that were consistent with artificial selection, including GPC4, which is known to be associated with growth rates in humans. Levels of diversity were not significantly different between autosomal and X-linked loci, providing no evidence for differential contributions of males and females to the domesticated gene pool. The structure of LD differed substantially within and among breeds. Within breeds, LD extends over large genomic distances. Markers separated by 400 kb typically showed r(2) higher than 0.2, and some LD extended up to 3,200 kb. Much less LD was found among breeds. This advantageous LD structure holds great promise for reducing the interval of association in future mapping studies.