Understanding metric-related pitfalls in image analysis validation.

Validation metrics are key for tracking scientific progress and bridging the current chasm between artificial intelligence research and its translation into practice. However, increasing evidence shows that, particularly in image analysis, metrics are often chosen inadequately. Although taking into account the individual strengths, weaknesses and limitations of validation metrics is a critical prerequisite to making educated choices, the relevant knowledge is currently scattered and poorly accessible to individual researchers. Based on a multistage Delphi process conducted by a multidisciplinary expert consortium as well as extensive community feedback, the present work provides a reliable and comprehensive common point of access to information on pitfalls related to validation metrics in image analysis. Although focused on biomedical image analysis, the addressed pitfalls generalize across application domains and are categorized according to a newly created, domain-agnostic taxonomy. The work serves to enhance global comprehension of a key topic in image analysis validation.

Biophysical characterization of calcium-binding and modulatory-domain dynamics in a pentameric ligand-gated ion channel.

Pentameric ligand-gated ion channels (pLGICs) perform electrochemical signal transduction in organisms ranging from bacteria to humans. Among the prokaryotic pLGICs, there is architectural diversity involving N-terminal domains (NTDs) not found in eukaryotic relatives, exemplified by the calcium-sensitive channel (DeCLIC) from a Desulfofustis deltaproteobacterium, which has an NTD in addition to the canonical pLGIC structure. Here, we have characterized the structure and dynamics of DeCLIC through cryoelectron microscopy (cryo-EM), small-angle neutron scattering (SANS), and molecular dynamics (MD) simulations. In the presence and absence of calcium, cryo-EM yielded structures with alternative conformations of the calcium-binding site. SANS profiles further revealed conformational diversity at room temperature beyond that observed in static structures, shown through MD to be largely attributable to rigid-body motions of the NTD relative to the protein core, with expanded and asymmetric conformations improving the fit of the SANS data. This work reveals the range of motion available to the DeCLIC NTD and calcium-binding site, expanding the conformational landscape of the pLGIC family. Further, these findings demonstrate the power of combining low-resolution scattering, high-resolution structural, and MD simulation data to elucidate interfacial interactions that are highly conserved in the pLGIC family.

Structural characterisation of α-synuclein-membrane interactions and the resulting aggregation using small angle scattering.

The presence of amyloid fibrils is a hallmark of several neurodegenerative diseases. Some amyloidogenic proteins, such as α-synuclein and amyloid ß, interact with lipids, and this interaction can strongly favour the formation of amyloid fibrils. In particular the primary nucleation step, i.e. the de novo formation of amyloid fibrils, has been shown to be accelerated by lipids. However, the exact mechanism of this acceleration is still mostly unclear. Here we use a range of scattering methods, such as dynamic light scattering (DLS) and small angle X-ray and neutron scattering (SAXS and SANS) to obtain structural information on the binding of α-synuclein to model membranes formed from negatively charged lipids and their co-assembly into amyloid fibrils. We find that the model membranes take an active role in the reaction. The binding of α synuclein to the model membranes immediately induces a major structural change in the lipid assembly, which leads to a break-up into small and mostly disc- or rod-like lipid-protein particles. This transition can be reversed by temperature changes or proteolytic protein removal. Incubation of the small lipid-α-synuclein particles for several hours, however, leads to amyloid fibril formation, whereby the lipids are incorporated into the amyloid fibrils.

Probing solution structure of the pentameric ligand-gated ion channel GLIC by small-angle neutron scattering.

Pentameric ligand-gated ion channels undergo subtle conformational cycling to control electrochemical signal transduction in many kingdoms of life. Several crystal structures have now been reported in this family, but the functional relevance of such models remains unclear. Here, we used small-angle neutron scattering (SANS) to probe ambient solution-phase properties of the pH-gated bacterial ion channel GLIC under resting and activating conditions. Data collection was optimized by inline paused-flow size-exclusion chromatography, and exchanging into deuterated detergent to hide the micelle contribution. Resting-state GLIC was the best-fit crystal structure to SANS curves, with no evidence for divergent mechanisms. Moreover, enhanced-sampling molecular-dynamics simulations enabled differential modeling in resting versus activating conditions, with the latter corresponding to an intermediate ensemble of both the extracellular and transmembrane domains. This work demonstrates state-dependent changes in a pentameric ion channel by SANS, an increasingly accessible method for macromolecular characterization with the coming generation of neutron sources.

Solution Structures of Two Different FRP-OCP Complexes as Revealed via SEC-SANS.

Photosynthetic organisms have established photoprotective mechanisms in order to dissipate excess light energy into heat, which is commonly known as non-photochemical quenching. Cyanobacteria utilize the orange carotenoid protein (OCP) as a high-light sensor and quencher to regulate the energy flow in the photosynthetic apparatus. Triggered by strong light, OCP undergoes conformational changes to form the active red state (OCPR). In many cyanobacteria, the back conversion of OCP to the dark-adapted state is assisted by the fluorescence recovery protein (FRP). However, the exact molecular events involving OCP and its interaction with FRP remain largely unraveled so far due to their metastability. Here, we use small-angle neutron scattering combined with size exclusion chromatography (SEC-SANS) to unravel the solution structures of FRP-OCP complexes using a compact mutant of OCP lacking the N-terminal extension (∆NTEOCPO) and wild-type FRP. The results are consistent with the simultaneous presence of stable 2:2 and 2:1 FRP-∆NTEOCPO complexes in solution, where the former complex type is observed for the first time. For both complex types, we provide ab initio low-resolution shape reconstructions and compare them to homology models based on available crystal structures. It is likely that both complexes represent intermediate states of the back conversion of OCP to its dark-adapted state in the presence of FRP, which are of transient nature in the photocycle of wild-type OCP. This study demonstrates the large potential of SEC-SANS in revealing the solution structures of protein complexes in polydisperse solutions that would otherwise be averaged, leading to unspecific results.

Small-angle x-ray and neutron scattering of MexR and its complex with DNA supports a conformational selection binding model.

In this work, we used small-angle x-ray and neutron scattering to reveal the shape of the protein-DNA complex of the Pseudomonas aeruginosa transcriptional regulator MexR, a member of the multiple antibiotics resistance regulator (MarR) family, when bound to one of its native DNA binding sites. Several MarR-like proteins, including MexR, repress the expression of efflux pump proteins by binding to DNA on regulatory sites overlapping with promoter regions. When expressed, efflux proteins self-assemble to form multiprotein complexes and actively expel highly toxic compounds out of the host organism. The mutational pressure on efflux-regulating MarR family proteins is high since deficient DNA binding leads to constitutive expression of efflux pumps and thereby supports acquired multidrug resistance. Understanding the functional outcome of such mutations and their effects on DNA binding has been hampered by the scarcity of structural and dynamic characterization of both free and DNA-bound MarR proteins. Here, we show how combined neutron and x-ray small-angle scattering of both states in solution support a conformational selection model that enhances MexR asymmetry in binding to one of its promoter-overlapping DNA binding sites.

ROOD-MRI: Benchmarking the robustness of deep learning segmentation models to out-of-distribution and corrupted data in MRI.

Deep artificial neural networks (DNNs) have moved to the forefront of medical image analysis due to their success in classification, segmentation, and detection challenges. A principal challenge in large-scale deployment of DNNs in neuroimage analysis is the potential for shifts in signal-to-noise ratio, contrast, resolution, and presence of artifacts from site to site due to variances in scanners and acquisition protocols. DNNs are famously susceptible to these distribution shifts in computer vision. Currently, there are no benchmarking platforms or frameworks to assess the robustness of new and existing models to specific distribution shifts in MRI, and accessible multi-site benchmarking datasets are still scarce or task-specific. To address these limitations, we propose ROOD-MRI: a novel platform for benchmarking the Robustness of DNNs to Out-Of-Distribution (OOD) data, corruptions, and artifacts in MRI. This flexible platform provides modules for generating benchmarking datasets using transforms that model distribution shifts in MRI, implementations of newly derived benchmarking metrics for image segmentation, and examples for using the methodology with new models and tasks. We apply our methodology to hippocampus, ventricle, and white matter hyperintensity segmentation in several large studies, providing the hippocampus dataset as a publicly available benchmark. By evaluating modern DNNs on these datasets, we demonstrate that they are highly susceptible to distribution shifts and corruptions in MRI. We show that while data augmentation strategies can substantially improve robustness to OOD data for anatomical segmentation tasks, modern DNNs using augmentation still lack robustness in more challenging lesion-based segmentation tasks. We finally benchmark U-Nets and vision transformers, finding robustness susceptibility to particular classes of transforms across architectures. The presented open-source platform enables generating new benchmarking datasets and comparing across models to study model design that results in improved robustness to OOD data and corruptions in MRI.

Human myelin proteolipid protein structure and lipid bilayer stacking.

The myelin sheath is an essential, multilayered membrane structure that insulates axons, enabling the rapid transmission of nerve impulses. The tetraspan myelin proteolipid protein (PLP) is the most abundant protein of compact myelin in the central nervous system (CNS). The integral membrane protein PLP adheres myelin membranes together and enhances the compaction of myelin, having a fundamental role in myelin stability and axonal support. PLP is linked to severe CNS neuropathies, including inherited Pelizaeus-Merzbacher disease and spastic paraplegia type 2, as well as multiple sclerosis. Nevertheless, the structure, lipid interaction properties, and membrane organization mechanisms of PLP have remained unidentified. We expressed, purified, and structurally characterized human PLP and its shorter isoform DM20. Synchrotron radiation circular dichroism spectroscopy and small-angle X-ray and neutron scattering revealed a dimeric, α-helical conformation for both PLP and DM20 in detergent complexes, and pinpoint structural variations between the isoforms and their influence on protein function. In phosphatidylcholine membranes, reconstituted PLP and DM20 spontaneously induced formation of multilamellar myelin-like membrane assemblies. Cholesterol and sphingomyelin enhanced the membrane organization but were not crucial for membrane stacking. Electron cryomicroscopy, atomic force microscopy, and X-ray diffraction experiments for membrane-embedded PLP/DM20 illustrated effective membrane stacking and ordered organization of membrane assemblies with a repeat distance in line with CNS myelin. Our results shed light on the 3D structure of myelin PLP and DM20, their structure-function differences, as well as fundamental protein-lipid interplay in CNS compact myelin.

A Convolutional Neural Network for Beamforming and Image Reconstruction in Passive Cavitation Imaging.

Convolutional neural networks (CNNs), initially developed for image processing applications, have recently received significant attention within the field of medical ultrasound imaging. In this study, passive cavitation imaging/mapping (PCI/PAM), which is used to map cavitation sources based on the correlation of signals across an array of receivers, is evaluated. Traditional reconstruction techniques in PCI, such as delay-and-sum, yield high spatial resolution at the cost of a substantial computational time. This results from the resource-intensive process of determining sensor weights for individual pixels in these methodologies. Consequently, the use of conventional algorithms for image reconstruction does not meet the speed requirements that are essential for real-time monitoring. Here, we show that a three-dimensional (3D) convolutional network can learn the image reconstruction algorithm for a 16×16 element matrix probe with a receive frequency ranging from 256 kHz up to 1.0 MHz. The network was trained and evaluated using simulated data representing point sources, resulting in the successful reconstruction of volumetric images with high sensitivity, especially for single isolated sources (100% in the test set). As the number of simultaneous sources increased, the network's ability to detect weaker intensity sources diminished, although it always correctly identified the main lobe. Notably, however, network inference was remarkably fast, completing the task in approximately 178 s for a dataset comprising 650 frames of 413 volume images with signal duration of 20µs. This processing speed is roughly thirty times faster than a parallelized implementation of the traditional time exposure acoustics algorithm on the same GPU device. This would open a new door for PCI application in the real-time monitoring of ultrasound ablation.

Connectivity of the corticostriatal and thalamostriatal systems in normal and parkinsonian states: An update.

The striatum receives abundant glutamatergic afferents from the cortex and thalamus. These inputs play a major role in the functions of the striatal neurons in normal conditions, and are significantly altered in pathological states, such as Parkinson's disease. This review summarizes the current knowledge of the connectivity of the corticostriatal and thalamostriatal pathways, with emphasis on the most recent advances in the field. We also discuss novel findings regarding structural changes in cortico- and thalamostriatal connections that occur in these connections as a consequence of striatal loss of dopamine in parkinsonism.

Quantitative single-cell analysis of immunofluorescence protein multiplex images illustrates biomarker spatial heterogeneity within breast cancer subtypes.

BACKGROUND: The extent of cellular heterogeneity in breast cancer could have potential impact on diagnosis and long-term outcome. However, pathology evaluation is limited to biomarker immunohistochemical staining and morphology of the bulk cancer. Inter-cellular heterogeneity of biomarkers is not usually assessed. As an initial evaluation of the extent of breast cancer cellular heterogeneity, we conducted quantitative and spatial imaging of Estrogen Receptor (ER), Progesterone Receptor (PR), Epidermal Growth Factor Receptor-2 (HER2), Ki67, TP53, CDKN1A (P21/WAF1), CDKN2A (P16INK4A), CD8 and CD20 of a tissue microarray (TMA) representing subtypes defined by St. Gallen surrogate classification. METHODS: Quantitative, single cell-based imaging was conducted using an Immunofluorescence protein multiplexing platform (MxIF) to study protein co-expression signatures and their spatial localization patterns. The range of MxIF intensity values of each protein marker was compared to the respective IHC score for the TMA core. Extent of heterogeneity in spatial neighborhoods was analyzed using co-occurrence matrix and Diversity Index measures. RESULTS: On the 101 cores from 59 cases studied, diverse expression levels and distributions were observed in MxIF measures of ER and PR among the hormonal receptor-positive tumor cores. As expected, Luminal A-like cancers exhibit higher proportions of cell groups that co-express ER and PR, while Luminal B-like (HER2-negative) cancers were composed of ER+, PR- groups. Proliferating cells defined by Ki67 positivity were mainly found in groups with PR-negative cells. Triple-Negative Breast Cancer (TNBC) exhibited the highest proliferative fraction and incidence of abnormal P53 and P16 expression. Among the tumors exhibiting P53 overexpression by immunohistochemistry, a group of TNBC was found with much higher MxIF-measured P53 signal intensity compared to HER2+, Luminal B-like and other TNBC cases. Densities of CD8 and CD20 cells were highest in HER2+ cancers. Spatial analysis demonstrated variability in heterogeneity in cellular neighborhoods in the cancer and the tumor microenvironment. CONCLUSIONS: Protein marker multiplexing and quantitative image analysis demonstrated marked heterogeneity in protein co-expression signatures and cellular arrangement within each breast cancer subtype. These refined descriptors of biomarker expressions and spatial patterns could be valuable in the development of more informative tools to guide diagnosis and treatment.

Temporal processing in the striatum: Interplay between midbrain dopamine neurons and striatal cholinergic interneurons.

There has been considerable progress in recent years toward understanding the neuronal mechanisms mediating time perception. Notably, the striatum and its dopamine (DA) input from the ventral midbrain are considered to be central for timing on the scale of hundreds of milliseconds and seconds. The cholinergic interneurons (ChIs) of the striatum provide an extensive local innervation, which closely interacts with striatal DA afferents. Both neuronal systems have been shown to influence synaptic plasticity to shape the transfer of information through the striatum. Given their cooperative role in regulating striatal output pathways, DA and cholinergic inputs may have distinct but complementary roles in timing processes. Electrophysiological recordings from behaving animals have provided evidence that responses of midbrain DA neurons and striatal tonically active neurons (TANs), presumed ChIs, to motivationally relevant events are sensitive to the predicted time of these events; namely, changes in neuronal activity are reduced or absent at times when events are more expected, indicating that temporal aspects of prediction play an important role in the responsiveness of these two neuronal systems. Recently, new findings have further suggested that DA neurons and cholinergic TANs are both involved in the ability to keep track of the elapsed time. These two systems appear to work in parallel in initiating the timing process at the beginning of an interval to be timed. It therefore appears that DA and ChI signaling could participate in striatal processing that is crucial for the control of timing behavior.

Structural and Biophysical Properties of Supercharged and Circularized Nanodiscs.

Nanodiscs based on membrane scaffold proteins (MSPs) and phospholipids are used as membrane mimics to stabilize membrane proteins in solution for structural and functional studies. Combining small-angle X-ray scattering (SAXS), differential scanning calorimetry (DSC), and time-resolved small-angle neutron scattering (TR-SANS), we characterized the structure and lipid bilayer properties of five different nanodiscs made with dimyristoylphosphatidylcholine and different MSPs varying in size, charge, and circularization. Our SAXS modeling showed that the structural parameters of the embedded lipids are all similar, irrespective of the MSP properties. DSC showed that the lipid packing is not homogeneous in the nanodiscs and that a 20 Å wide boundary layer of lipids with perturbed packing is located close to the MSP, while the packing of central lipids is tighter than in large unilamellar vesicles. Finally, TR-SANS showed that lipid exchange rates in nanodiscs decrease with increasing nanodisc size and are lower for the nanodiscs made with supercharged MSPs compared to conventional nanodiscs. Altogether, the results provide a thorough biophysical understanding of the nanodisc as a model membrane system, which is important in order to carry out and interpret experiments on membrane proteins embedded in such systems.

Interpenetrated biosurfactant-silk fibroin networks - a SANS study.

Silk fibroin (SF) based hydrogels have been exploited for years for their inherent biocompatibility and favorable mechanical properties which makes them interesting for biotechnology applications. In this study we investigate silk based composite hydrogels where pH-sensitive, anionic biosurfactant assemblies (sophorolipids SL-C18 : 1 and SL-C18 : 0), are employed to improve the present properties of SF. Results suggest that the presence of SL surfactant assemblies leads to faster gelling of SF by accelerating the refolding from random coil to ß-sheet as shown by infrared and UV-visible spectroscopy. Small angle neutron scattering (SANS) including contrast matching studies show that SF and SL assemblies coexist in a fibrillary network that is, in the case of SL-C18 : 0, interpenetrating. The resulting overall network structure in composite gels is slightly more affected by SL-C18 : 1 than by SL-C18 : 0, whereas the structure of both SF and surfactant assemblies remains unchanged. No disassembly of SL surfactant structures is observed, which gives a new perspective on SF-surfactant interactions. The hydrophobic effect within SF is favored in the presence of SL, leading to faster refolding of SF into ß-sheet conformation. The presented composite gels, being an interpenetrating network of which one compound (SL-C18 : 0) can be tweaked by pH, open an interesting option towards improved workability and stimuli responsive mechanical properties of SF based hydrogels with possible applications in controlled cell culture and tissue engineering or drug delivery. The presented SANS analysis approach has the potential to be expanded to other protein-surfactant systems and composite hydrogels.

The Impact of Redox, Hydrolysis and Dehydration Chemistry on the Structural and Magnetic Properties of Magnetoferritin Prepared in Variable Thermal Conditions.

Ferritin, a spherically shaped protein complex, is responsible for iron storage in bacteria, plants, animals, and humans. Various ferritin iron core compositions in organisms are associated with specific living requirements, health state, and different biochemical roles of ferritin isomers. Magnetoferritin, a synthetic ferritin derivative, serves as an artificial model system of unusual iron phase structures found in humans. We present the results of a complex structural study of magnetoferritins prepared by controlled in vitro synthesis. Using various complementary methods, it was observed that manipulation of the synthesis technology can improve the physicochemical parameters of the system, which is useful in applications. Thus, a higher synthesis temperature leads to an increase in magnetization due to the formation of the magnetite phase. An increase in the iron loading factor has a more pronounced impact on the protein shell structure in comparison with the pH of the aqueous medium. On the other hand, a higher loading factor at physiological temperature enhances the formation of an amorphous phase instead of magnetite crystallization. It was confirmed that the iron-overloading effect alone (observed during pathological events) cannot contribute to the formation of magnetite.

Observing Protein Degradation by the PAN-20S Proteasome by Time-Resolved Neutron Scattering.

The proteasome is a key player of regulated protein degradation in all kingdoms of life. Although recent atomic structures have provided snapshots on a number of conformations, data on substrate states and populations during the active degradation process in solution remain scarce. Here, we use time-resolved small-angle neutron scattering of a deuterium-labeled GFPssrA substrate and an unlabeled archaeal PAN-20S system to obtain direct structural information on substrate states during ATP-driven unfolding and subsequent proteolysis in solution. We find that native GFPssrA structures are degraded in a biexponential process, which correlates strongly with ATP hydrolysis, the loss of fluorescence, and the buildup of small oligopeptide products. Our solution structural data support a model in which the substrate is directly translocated from PAN into the 20S proteolytic chamber, after a first, to our knowledge, successful unfolding process that represents a point of no return and thus prevents dissociation of the complex and the release of harmful, aggregation-prone products.

Structural Studies of Overlapping Dinucleosomes in Solution.

An overlapping dinucleosome (OLDN) is a structure composed of one hexasome and one octasome and appears to be formed through nucleosome collision promoted by nucleosome remodeling factor(s). In this study, the solution structure of the OLDN was investigated through the integration of small-angle x-ray and neutron scattering (SAXS and SANS, respectively), computer modeling, and molecular dynamics simulations. Starting from the crystal structure, we generated a conformational ensemble based on normal mode analysis and searched for the conformations that reproduced the SAXS and SANS scattering curves well. We found that inclusion of histone tails, which are not observed in the crystal structure, greatly improved model quality. The obtained structural models suggest that OLDNs adopt a variety of conformations stabilized by histone tails situated at the interface between the hexasome and octasome, simultaneously binding to both the hexasomal and octasomal DNA. In addition, our models define a possible direction for the conformational changes or dynamics, which may provide important information that furthers our understanding of the role of chromatin dynamics in gene regulation.

Interdomain Flexibility within NADPH Oxidase Suggested by SANS Using LMNG Stealth Carrier.

Small angle neutron scattering (SANS) provides a method to obtain important low-resolution information for integral membrane proteins (IMPs), challenging targets for structural determination. Specific deuteration furnishes a "stealth" carrier for the solubilized IMP. We used SANS to determine a structural envelope of SpNOX, the Streptococcus pneumoniae NADPH oxidase (NOX), a prokaryotic model system for exploring structure and function of eukaryotic NOXes. SpNOX was solubilized in the detergent lauryl maltose neopentyl glycol, which provides optimal SpNOX stability and activity. Using deuterated solvent and protein, the lauryl maltose neopentyl glycol was experimentally undetected in SANS. This affords a cost-effective SANS approach for obtaining novel structural information on IMPs. Combining SANS data with molecular modeling provided a first, to our knowledge, structural characterization of an entire NOX enzyme. It revealed a distinctly less compact structure than that predicted from the docking of homologous crystal structures of the separate transmembrane and dehydrogenase domains, consistent with a flexible linker connecting the two domains.

How the central domain of dystrophin acts to bridge F-actin to sarcolemmal lipids.

Dystrophin is a large intracellular protein that prevents sarcolemmal ruptures by providing a mechanical link between the intracellular actin cytoskeleton and the transmembrane dystroglycan complex. Dystrophin deficiency leads to the severe muscle wasting disease Duchenne Muscular Dystrophy and the milder allelic variant, Becker Muscular Dystrophy (DMD and BMD). Previous work has shown that concomitant interaction of the actin binding domain 2 (ABD2) comprising spectrin like repeats 11 to 15 (R11-15) of the central domain of dystrophin, with both actin and membrane lipids, can greatly increase membrane stiffness. Based on a combination of SAXS and SANS measurements, mass spectrometry analysis of cross-linked complexes and interactive low-resolution simulations, we explored in vitro the molecular properties of dystrophin that allow the formation of ABD2-F-actin and ABD2-membrane model complexes. In dystrophin we identified two subdomains interacting with F-actin, one located in R11 and a neighbouring region in R12 and another one in R15, while a single lipid binding domain was identified at the C-terminal end of R12. Relative orientations of the dystrophin central domain with F-actin and a membrane model were obtained from docking simulation under experimental constraints. SAXS-based models were then built for an extended central subdomain from R4 to R19, including ABD2. Overall results are compatible with a potential F-actin/dystrophin/membrane lipids ternary complex. Our description of this selected part of the dystrophin associated complex bridging muscle cell membrane and cytoskeleton opens the way to a better understanding of how cell muscle scaffolding is maintained through this essential protein.

Hippocampal segmentation for brains with extensive atrophy using three-dimensional convolutional neural networks.

Hippocampal volumetry is a critical biomarker of aging and dementia, and it is widely used as a predictor of cognitive performance; however, automated hippocampal segmentation methods are limited because the algorithms are (a) not publicly available, (b) subject to error with significant brain atrophy, cerebrovascular disease and lesions, and/or (c) computationally expensive or require parameter tuning. In this study, we trained a 3D convolutional neural network using 259 bilateral manually delineated segmentations collected from three studies, acquired at multiple sites on different scanners with variable protocols. Our training dataset consisted of elderly cases difficult to segment due to extensive atrophy, vascular disease, and lesions. Our algorithm, (HippMapp3r), was validated against four other publicly available state-of-the-art techniques (HippoDeep, FreeSurfer, SBHV, volBrain, and FIRST). HippMapp3r outperformed the other techniques on all three metrics, generating an average Dice of 0.89 and a correlation coefficient of 0.95. It was two orders of magnitude faster than some of the tested techniques. Further validation was performed on 200 subjects from two other disease populations (frontotemporal dementia and vascular cognitive impairment), highlighting our method's low outlier rate. We finally tested the methods on real and simulated "clinical adversarial" cases to study their robustness to corrupt, low-quality scans. The pipeline and models are available at: https://hippmapp3r.readthedocs.ioto facilitate the study of the hippocampus in large multisite studies.