Graph-in-Graph (GiG): Learning interpretable latent graphs in non-Euclidean domain for biological and healthcare applications.

Graphs are a powerful tool for representing and analyzing unstructured, non-Euclidean data ubiquitous in the healthcare domain. Two prominent examples are molecule property prediction and brain connectome analysis. Importantly, recent works have shown that considering relationships between input data samples has a positive regularizing effect on the downstream task in healthcare applications. These relationships are naturally modeled by a (possibly unknown) graph structure between input samples. In this work, we propose Graph-in-Graph (GiG), a neural network architecture for protein classification and brain imaging applications that exploits the graph representation of the input data samples and their latent relation. We assume an initially unknown latent-graph structure between graph-valued input data and propose to learn a parametric model for message passing within and across input graph samples, end-to-end along with the latent structure connecting the input graphs. Further, we introduce a Node Degree Distribution Loss (NDDL) that regularizes the predicted latent relationships structure. This regularization can significantly improve the downstream task. Moreover, the obtained latent graph can represent patient population models or networks of molecule clusters, providing a level of interpretability and knowledge discovery in the input domain, which is of particular value in healthcare.

ToxNet: an artificial intelligence designed for decision support for toxin prediction.

BACKGROUND: Artificial intelligences (AIs) are emerging in the field of medical informatics in many areas. They are mostly used for diagnosis support in medical imaging but have potential uses in many other fields of medicine where large datasets are available. AIM: To develop an artificial intelligence (AI) "ToxNet", a machine-learning based computer-aided diagnosis (CADx) system, which aims to predict poisons based on patient's symptoms and metadata from our Poison Control Center (PCC) data. To prove its accuracy and compare it against medical doctors (MDs). METHODS: The CADx system was developed and trained using data from 781,278 calls recorded in our PCC database from 2001 to 2019. All cases were mono-intoxications. Patient symptoms and meta-information (e.g., age group, sex, etiology, toxin point of entry, weekday, etc.) were provided. In the pilot phase, the AI was trained on 10 substances, the AI's prediction was compared to naïve matching, literature matching, a multi-layer perceptron (MLP), and the graph attention network (GAT). The trained AI's accuracy was then compared to 10 medical doctors in an individual and in an identical dataset. The dataset was then expanded to 28 substances and the predictions and comparisons repeated. RESULTS: In the pilot, the prediction performance in a set of 8995 patients with 10 substances was 0.66 ± 0.01 (F1 micro score). Our CADx system was significantly superior to naïve matching, literature matching, MLP, and GAT (p < 0.005). It outperformed our physicians experienced in clinical toxicology in the individual and identical dataset. In the extended dataset, our CADx system was able to predict the correct toxin in a set of 36,033 patients with 28 substances with an overall performance of 0.27 ± 0.01 (F1 micro score), also significantly superior to naïve matching, literature matching, MLP, and GAT. It also outperformed our MDs. CONCLUSION: Our AI trained on a large PCC database works well for poison prediction in these experiments. With further research, it might become a valuable aid for physicians in predicting unknown substances and might be the first step into AI use in PCCs.

Differentiable Graph Module (DGM) for Graph Convolutional Networks.

Graph deep learning has recently emerged as a powerful ML concept allowing to generalize successful deep neural architectures to non-euclidean structured data. Such methods have shown promising results on a broad spectrum of applications ranging from social science, biomedicine, and particle physics to computer vision, graphics, and chemistry. One of the limitations of the majority of current graph neural network architectures is that they are often restricted to the transductive setting and rely on the assumption that the underlying graph is known and fixed. Often, this assumption is not true since the graph may be noisy, or partially and even completely unknown. In such cases, it would be helpful to infer the graph directly from the data, especially in inductive settings where some nodes were not present in the graph at training time. Furthermore, learning a graph may become an end in itself, as the inferred structure may provide complementary insights next to the downstream task. In this paper, we introduce Differentiable Graph Module (DGM), a learnable function that predicts edge probabilities in the graph which are optimal for the downstream task. DGM can be combined with convolutional graph neural network layers and trained in an end-to-end fashion. We provide an extensive evaluation of applications from the domains of healthcare (disease prediction), brain imaging (age prediction), computer graphics (3D point cloud segmentation), and computer vision (zero-shot learning). We show that our model provides a significant improvement over baselines both in transductive and inductive settings and achieves state-of-the-art results.

IE-Vnet: Deep Learning-Based Segmentation of the Inner Ear's Total Fluid Space.

Background: In-vivo MR-based high-resolution volumetric quantification methods of the endolymphatic hydrops (ELH) are highly dependent on a reliable segmentation of the inner ear's total fluid space (TFS). This study aimed to develop a novel open-source inner ear TFS segmentation approach using a dedicated deep learning (DL) model. Methods: The model was based on a V-Net architecture (IE-Vnet) and a multivariate (MR scans: T1, T2, FLAIR, SPACE) training dataset (D1, 179 consecutive patients with peripheral vestibulocochlear syndromes). Ground-truth TFS masks were generated in a semi-manual, atlas-assisted approach. IE-Vnet model segmentation performance, generalizability, and robustness to domain shift were evaluated on four heterogenous test datasets (D2-D5, n = 4 × 20 ears). Results: The IE-Vnet model predicted TFS masks with consistently high congruence to the ground-truth in all test datasets (Dice overlap coefficient: 0.9 ± 0.02, Hausdorff maximum surface distance: 0.93 ± 0.71 mm, mean surface distance: 0.022 ± 0.005 mm) without significant difference concerning side (two-sided Wilcoxon signed-rank test, p>0.05), or dataset (Kruskal-Wallis test, p>0.05; post-hoc Mann-Whitney U, FDR-corrected, all p>0.2). Prediction took 0.2 s, and was 2,000 times faster than a state-of-the-art atlas-based segmentation method. Conclusion: IE-Vnet TFS segmentation demonstrated high accuracy, robustness toward domain shift, and rapid prediction times. Its output works seamlessly with a previously published open-source pipeline for automatic ELS segmentation. IE-Vnet could serve as a core tool for high-volume trans-institutional studies of the inner ear. Code and pre-trained models are available free and open-source under https://github.com/pydsgz/IEVNet.

Persistent horizontal and vertical, MR-induced nystagmus in resting state Human Connectome Project data.

OBJECTIVE: Strong magnetic fields from magnetic resonance (MR) scanners induce a Lorentz force that contributes to vertigo and persistent nystagmus. Prior studies have reported a predominantly horizontal direction for healthy subjects in a 7 Tesla (T) MR scanner, with slow phase velocity (SPV) dependent on head orientation. Less is known about vestibular signal behavior for subjects in a weaker, 3T magnetic field, the standard strength used in the Human Connectome Project (HCP). The purpose of this study is to characterize the form and magnitude of nystagmus induced at 3T. METHODS: Forty-two subjects were studied after being introduced head-first, supine into a Siemens Prisma 3T scanner. Eye movements were recorded in four separate acquisitions over 20 min. A biometric eye model was fitted to the recordings to derive rotational eye position and then SPV. An anatomical template of the semi-circular canals was fitted to the T2 anatomical image from each subject, and used to derive the angle of the B0 magnetic field with respect to the vestibular apparatus. RESULTS: Recordings from 37 subjects yielded valid measures of eye movements. The population-mean SPV ± SD for the horizontal component was -1.38 ± 1.27 deg/sec, and vertical component was -0.93 ± 1.44 deg/sec, corresponding to drift movement in the rightward and downward direction. Although there was substantial inter-subject variability, persistent nystagmus was present in half of subjects with no significant adaptation over the 20 min scanning period. The amplitude of vertical drift was correlated with the roll angle of the vestibular system, with a non-zero vertical SPV present at a 0 degree roll. INTERPRETATION: Non-habituating vestibular signals of varying amplitude are present in resting state data collected at 3T.

Dosimetric impact of deep learning-based CT auto-segmentation on radiation therapy treatment planning for prostate cancer.

BACKGROUND: The evaluation of automatic segmentation algorithms is commonly performed using geometric metrics. An analysis based on dosimetric parameters might be more relevant in clinical practice but is often lacking in the literature. The aim of this study was to investigate the impact of state-of-the-art 3D U-Net-generated organ delineations on dose optimization in radiation therapy (RT) for prostate cancer patients. METHODS: A database of 69 computed tomography images with prostate, bladder, and rectum delineations was used for single-label 3D U-Net training with dice similarity coefficient (DSC)-based loss. Volumetric modulated arc therapy (VMAT) plans have been generated for both manual and automatic segmentations with the same optimization settings. These were chosen to give consistent plans when applying perturbations to the manual segmentations. Contours were evaluated in terms of DSC, average and 95% Hausdorff distance (HD). Dose distributions were evaluated with the manual segmentation as reference using dose volume histogram (DVH) parameters and a 3%/3 mm gamma-criterion with 10% dose cut-off. A Pearson correlation coefficient between DSC and dosimetric metrics, i.e. gamma index and DVH parameters, has been calculated. RESULTS: 3D U-Net-based segmentation achieved a DSC of 0.87 (0.03) for prostate, 0.97 (0.01) for bladder and 0.89 (0.04) for rectum. The mean and 95% HD were below 1.6 (0.4) and below 5 (4) mm, respectively. The DVH parameters, V[Formula: see text] for the bladder and V[Formula: see text] for the rectum, showed agreement between dose distributions within [Formula: see text] and [Formula: see text], respectively. The D[Formula: see text] and V[Formula: see text], for prostate and its 3 mm expansion (surrogate clinical target volume) showed agreement with the reference dose distribution within 2% and 3 Gy with the exception of one case. The average gamma pass-rate was 85%. The comparison between geometric and dosimetric metrics showed no strong statistically significant correlation. CONCLUSIONS: The 3D U-Net developed for this work achieved state-of-the-art geometrical performance. Analysis based on clinically relevant DVH parameters of VMAT plans demonstrated neither excessive dose increase to OARs nor substantial under/over-dosage of the target in all but one case. Yet the gamma analysis indicated several cases with low pass rates. The study highlighted the importance of adding dosimetric analysis to the standard geometric evaluation.

Physics-aware learning and domain-specific loss design in ophthalmology.

The human cataract, a developing opacification of the human eye lens, currently constitutes the world's most frequent cause for blindness. As a result, cataract surgery has become the most frequently performed ophthalmic surgery in the world. By removing the human lens and replacing it with an artificial intraocular lens (IOL), the optical system of the eye is restored. In order to receive a good refractive result, the IOL specifications, especially the refractive power, have to be determined precisely prior to surgery. In the last years, there has been a body of work to perform this prediction by using biometric information extracted from OCT imaging data, recently also by machine learning (ML) methods. Approaches so far consider only biometric information or physical modelling, but provide no effective combination, while often also neglecting IOL geometry. Additionally, ML on small data sets without sufficient domain coverage can be challenging. To solve these issues, we propose OpticNet, a novel optical refraction network based on an unsupervised, domain-specific loss function that explicitly incorporates physical information into the network. By providing a precise and differentiable light propagation eye model, physical gradients following the eye optics are backpropagated into the network. We further propose a new transfer learning procedure, which allows the unsupervised pre-training on the optical model and fine-tuning of the network on small amounts of surgical patient data. We show that our method outperforms the current state of the art on five OCT-image based data sets, provides better domain coverage within its predictions, and achieves better physical consistency.

A comparative study of the status of supportive-palliative care provision in Iran and selected countries: Strengths and weaknesses.

BACKGROUND: Terminally, illnesses such as cancer, AIDS, dementia, and advanced heart disease will require special supportive and palliative care, although a few numbers of these patients are provided with these services. The aim of the present study was to perform a comparative study of supportive-palliative care provision in selected countries. MATERIALS AND METHODS: This research was a descriptive comparative study that its research population was the frameworks of palliative and supportive care provision in Egypt, Turkey, America, Australia, Canada, the Netherlands, and China. These frameworks were compared across six dimensions of service receivers, financing, providers, service provider centers, type of services provided, and training. Data collection tool has included the checklist and information sources, documents, evidence, articles, books, and journals collected through the Internet and organizations related to the health information of selected countries and by the library search. Data were investigated and analyzed using the data collection tool and checklists. FINDINGS: The findings showed that the developed countries having decentralized trusteeship structure had a more favorable status in palliative and supportive care provision. The type of services provided was a combination of mental, psychological, social, spiritual, financial, and physical and communication services. Provider centers included hospital, the elderly, and cancer and charity centers. CONCLUSION: Regarding the investigation and recognition of the status of supportive-palliative care provision, it was observed that the provision of these services was a concern of the selected countries, but they did not have a defined model or pattern to provide these services. Therefore, it is suggested that each country takes a step to redesign and define frameworks and structures in the evolution of supportive-palliative cares in accordance with the particular conditions, indigenous culture, religion, and other effective cases of that country and pays special attention to the role and position of supportive-palliative cares.

Investigating the challenges and barriers of palliative care delivery in Iran and the World: A systematic review study.

Palliative care and terminal patients care centers have an important role in improving the physical and psychological state of the patient and their families and increasing their satisfaction and care providers. A literature search of online databases (PubMed, Scopus, Web of science, Cochrane library, and Google Scholar) was searched from January 1, 2000, to the end of April 2019, by using the appropriate English keywords. Furthermore, IranMedex, Barkat, and Magiran databases were searched for the Persian articles. We used Standards for Reporting Qualitative Research checklist to evaluate the articles quality. From 1328 articles, 166 were reviewed in depth with 13 satisfying our inclusion criteria. The findings of this study revealed a wide range of barriers and challenges to palliative care delivery. The identified barriers were: Cultural, social and organizational barriers, lack of resources, equipment and financing, attitudes and cultures, barriers related to the patient and the patient's family, related barriers providers, time and money, education, communication challenges, policies, insurance problems, safety, and crisis management. The results of the studies showed that there are various barriers and challenges such as economic, cultural, social, organizational, and communication related to palliative care. Given the identified barriers and challenges, it is suggested that to improve the delivery of palliative care, the health system policy-makers and planners consider a resource-appropriate and culturally appropriate framework for palliative care delivery.

Using Base-ml to Learn Classification of Common Vestibular Disorders on DizzyReg Registry Data.

Background: Multivariable analyses (MVA) and machine learning (ML) applied on large datasets may have a high potential to provide clinical decision support in neuro-otology and reveal further avenues for vestibular research. To this end, we build base-ml, a comprehensive MVA/ML software tool, and applied it to three increasingly difficult clinical objectives in differentiation of common vestibular disorders, using data from a large prospective clinical patient registry (DizzyReg). Methods: Base-ml features a full MVA/ML pipeline for classification of multimodal patient data, comprising tools for data loading and pre-processing; a stringent scheme for nested and stratified cross-validation including hyper-parameter optimization; a set of 11 classifiers, ranging from commonly used algorithms like logistic regression and random forests, to artificial neural network models, including a graph-based deep learning model which we recently proposed; a multi-faceted evaluation of classification metrics; tools from the domain of "Explainable AI" that illustrate the input distribution and a statistical analysis of the most important features identified by multiple classifiers. Results: In the first clinical task, classification of the bilateral vestibular failure (N = 66) vs. functional dizziness (N = 346) was possible with a classification accuracy ranging up to 92.5% (Random Forest). In the second task, primary functional dizziness (N = 151) vs. secondary functional dizziness (following an organic vestibular syndrome) (N = 204), was classifiable with an accuracy ranging from 56.5 to 64.2% (k-nearest neighbors/logistic regression). The third task compared four episodic disorders, benign paroxysmal positional vertigo (N = 134), vestibular paroxysmia (N = 49), Menière disease (N = 142) and vestibular migraine (N = 215). Classification accuracy ranged between 25.9 and 50.4% (Naïve Bayes/Support Vector Machine). Recent (graph-) deep learning models classified well in all three tasks, but not significantly better than more traditional ML methods. Classifiers reliably identified clinically relevant features as most important toward classification. Conclusion: The three clinical tasks yielded classification results that correlate with the clinical intuition regarding the difficulty of diagnosis. It is favorable to apply an array of MVA/ML algorithms rather than a single one, to avoid under-estimation of classification accuracy. Base-ml provides a systematic benchmarking of classifiers, with a standardized output of MVA/ML performance on clinical tasks. To alleviate re-implementation efforts, we provide base-ml as an open-source tool for the community.

Simultaneous imputation and classification using Multigraph Geometric Matrix Completion (MGMC): Application to neurodegenerative disease classification.

Large-scale population-based studies in medicine are a key resource towards better diagnosis, monitoring, and treatment of diseases. They also serve as enablers of clinical decision support systems, in particular computer-aided diagnosis (CADx) using machine learning (ML). Numerous ML approaches for CADx have been proposed in literature. However, these approaches assume feature-complete data, which is often not the case in clinical data. To account for missing data, incomplete data samples are either removed or imputed, which could lead to data bias and may negatively affect classification performance. As a solution, we propose an end-to-end learning of imputation and disease prediction of incomplete medical datasets via Multi-graph Geometric Matrix Completion (MGMC). MGMC uses multiple recurrent graph convolutional networks, where each graph represents an independent population model based on a key clinical meta-feature like age, sex, or cognitive function. Graph signal aggregation from local patient neighborhoods, combined with multi-graph signal fusion via self-attention, has a regularizing effect on both matrix reconstruction and classification performance. Our proposed approach is able to impute class relevant features as well as perform accurate and robust classification on two publicly available medical datasets. We empirically show the superiority of our proposed approach in terms of classification and imputation performance when compared with state-of-the-art approaches. MGMC enables disease prediction in multimodal and incomplete medical datasets. These findings could serve as baseline for future CADx approaches which utilize incomplete datasets.

Metabolic connectivity-based single subject classification by multi-regional linear approximation in the rat.

Metabolic connectivity patterns on the basis of [18F]-FDG positron emission tomography (PET) are used to depict complex cerebral network alterations in different neurological disorders and therefore may have the potential to support diagnostic decisions. In this study, we established a novel statistical classification method taking advantage of differential time-dependent states of whole-brain metabolic connectivity following unilateral labyrinthectomy (UL) in the rat and explored its classification accuracy. The dataset consisted of repeated [18F]-FDG PET measurements at baseline and 1, 3, 7, and 15 days (= maximum of 5 classes) after UL with 17 rats per measurement day. Classification in different stages after UL was performed by determining connectivity patterns for the different classes by Pearson's correlation between uptake values in atlas-based segmented brain regions. Connections were fitted with a linear function, with which different thresholds on the correlation coefficient (r = [0.5, 0.85]) were investigated. Rats were classified by determining the congruence of their PET uptake pattern with the fitted connectivity patterns in the classes. Overall, the classification accuracy with this method was 84.3% for 3 classes, 75.0% for 4 classes, and 54.1% for 5 classes and outperformed random classification as well as machine learning classification on the same dataset. The optimal classification thresholds of the correlation coefficient and distance-to-fit were found to be |r| > 0.65 and d = 4 when using Siegel's slope estimator for fitting. This connectivity-based classification method can compete with machine learning classification and may have methodological advantages when applied to support PET-based diagnostic decisions in neurological network disorders (such as neurodegenerative syndromes).

IE-Map: a novel in-vivo atlas and template of the human inner ear.

Brain atlases and templates are core tools in scientific research with increasing importance also in clinical applications. Advances in neuroimaging now allowed us to expand the atlas domain to the vestibular and auditory organ, the inner ear. In this study, we present IE-Map, an in-vivo template and atlas of the human labyrinth derived from multi-modal high-resolution magnetic resonance imaging (MRI) data, in a fully non-invasive manner without any contrast agent or radiation. We reconstructed a common template from 126 inner ears (63 normal subjects) and annotated it with 94 established landmarks and semi-automatic segmentations of all relevant macroscopic vestibular and auditory substructures. We validated the atlas by comparing MRI templates to a novel CT/micro-CT atlas, which we reconstructed from 21 publicly available post-mortem images of the bony labyrinth. Templates in MRI and micro-CT have a high overlap, and several key anatomical measures of the bony labyrinth in IE-Map are in line with micro-CT literature of the inner ear. A quantitative substructural analysis based on the new template, revealed a correlation of labyrinth parameters with total intracranial volume. No effects of gender or laterality were found. We provide the validated templates, atlas segmentations, surface meshes and landmark annotations as open-access material, to provide neuroscience researchers and clinicians in neurology, neurosurgery, and otorhinolaryngology with a widely applicable tool for computational neuro-otology.

Modern machine-learning can support diagnostic differentiation of central and peripheral acute vestibular disorders.

BACKGROUND: Diagnostic classification of central vs. peripheral etiologies in acute vestibular disorders remains a challenge in the emergency setting. Novel machine-learning methods may help to support diagnostic decisions. In the current study, we tested the performance of standard and machine-learning approaches in the classification of consecutive patients with acute central or peripheral vestibular disorders. METHODS: 40 Patients with vestibular stroke (19 with and 21 without acute vestibular syndrome (AVS), defined by the presence of spontaneous nystagmus) and 68 patients with peripheral AVS due to vestibular neuritis were recruited in the emergency department, in the context of the prospective EMVERT trial (EMergency VERTigo). All patients received a standardized neuro-otological examination including videooculography and posturography in the acute symptomatic stage and an MRI within 7 days after symptom onset. Diagnostic performance of state-of-the-art scores, such as HINTS (Head Impulse, gaze-evoked Nystagmus, Test of Skew) and ABCD2 (Age, Blood, Clinical features, Duration, Diabetes), for the differentiation of vestibular stroke vs. peripheral AVS was compared to various machine-learning approaches: (i) linear logistic regression (LR), (ii) non-linear random forest (RF), (iii) artificial neural network, and (iv) geometric deep learning (Single/MultiGMC). A prospective classification was simulated by ten-fold cross-validation. We analyzed whether machine-estimated feature importances correlate with clinical experience. RESULTS: Machine-learning methods (e.g., MultiGMC) outperform univariate scores, such as HINTS or ABCD2, for differentiation of all vestibular strokes vs. peripheral AVS (MultiGMC area-under-the-curve (AUC): 0.96 vs. HINTS/ABCD2 AUC: 0.71/0.58). HINTS performed similarly to MultiGMC for vestibular stroke with AVS (AUC: 0.86), but more poorly for vestibular stroke without AVS (AUC: 0.54). Machine-learning models learn to put different weights on particular features, each of which is relevant from a clinical viewpoint. Established non-linear machine-learning methods like RF and linear methods like LR are less powerful classification models (AUC: 0.89 vs. 0.62). CONCLUSIONS: Established clinical scores (such as HINTS) provide a valuable baseline assessment for stroke detection in acute vestibular syndromes. In addition, machine-learning methods may have the potential to increase sensitivity and selectivity in the establishment of a correct diagnosis.

Analyzing the co-localization of substantia nigra hyper-echogenicities and iron accumulation in Parkinson's disease: A multi-modal atlas study with transcranial ultrasound and MRI.

BACKGROUND: Transcranial B-mode sonography (TCS) can detect hyperechogenic speckles in the area of the substantia nigra (SN) in Parkinson's disease (PD). These speckles correlate with iron accumulation in the SN tissue, but an exact volumetric localization in and around the SN is still unknown. Areas of increased iron content in brain tissue can be detected in vivo with magnetic resonance imaging, using quantitative susceptibility mapping (QSM). METHODS: In this work, we i) acquire, co-register and transform TCS and QSM imaging from a cohort of 23 PD patients and 27 healthy control subjects into a normalized atlas template space and ii) analyze and compare the 3D spatial distributions of iron accumulation in the midbrain, as detected by a signal increase (TCS+ and QSM+) in both modalities. RESULTS: We achieved sufficiently accurate intra-modal target registration errors (TRE<1 mm) for all MRI volumes and multi-modal TCS-MRI co-localization (TRE<4 mm) for 66.7% of TCS scans. In the caudal part of the midbrain, enlarged TCS+ and QSM+ areas were located within the SN pars compacta in PD patients in comparison to healthy controls. More cranially, overlapping TCS+ and QSM+ areas in PD subjects were found in the area of the ventral tegmental area (VTA). CONCLUSION: Our findings are concordant with several QSM-based studies on iron-related alterations in the area SN pars compacta. They substantiate that TCS+ is an indicator of iron accumulation in Parkinson's disease within and in the vicinity of the SN. Furthermore, they are in favor of an involvement of the VTA and thereby the mesolimbic system in Parkinson's disease.

Towards computerized diagnosis of neurological stance disorders: data mining and machine learning of posturography and sway.

We perform classification, ranking and mapping of body sway parameters from static posturography data of patients using recent machine-learning and data-mining techniques. Body sway is measured in 293 individuals with the clinical diagnoses of acute unilateral vestibulopathy (AVS, n = 49), distal sensory polyneuropathy (PNP, n = 12), anterior lobe cerebellar atrophy (CA, n = 48), downbeat nystagmus syndrome (DN, n = 16), primary orthostatic tremor (OT, n = 25), Parkinson's disease (PD, n = 27), phobic postural vertigo (PPV n = 59) and healthy controls (HC, n = 57). We classify disorders and rank sway features using supervised machine learning. We compute a continuous, human-interpretable 2D map of stance disorders using t-stochastic neighborhood embedding (t-SNE). Classification of eight diagnoses yielded 82.7% accuracy [95% CI (80.9%, 84.5%)]. Five (CA, PPV, AVS, HC, OT) were classified with a mean sensitivity and specificity of 88.4% and 97.1%, while three (PD, PNP, and DN) achieved a mean sensitivity of 53.7%. The most discriminative stance condition was ranked as "standing on foam-rubber, eyes closed". Mapping of sway path features into 2D space revealed clear clusters among CA, PPV, AVS, HC and OT subjects. We confirm previous claims that machine learning can aid in classification of clinical sway patterns measured with static posturography. Given a standardized, long-term acquisition of quantitative patient databases, modern machine learning and data analysis techniques help in visualizing, understanding and utilizing high-dimensional sensor data from clinical routine.

DeepVOG: Open-source pupil segmentation and gaze estimation in neuroscience using deep learning.

BACKGROUND: A prerequisite for many eye tracking and video-oculography (VOG) methods is an accurate localization of the pupil. Several existing techniques face challenges in images with artifacts and under naturalistic low-light conditions, e.g. with highly dilated pupils. NEW METHOD: For the first time, we propose to use a fully convolutional neural network (FCNN) for segmentation of the whole pupil area, trained on 3946 VOG images hand-annotated at our institute. We integrate the FCNN into DeepVOG, along with an established method for gaze estimation from elliptical pupil contours, which we improve upon by considering our FCNN's segmentation confidence measure. RESULTS: The FCNN output simultaneously enables us to perform pupil center localization, elliptical contour estimation and blink detection, all with a single network and with an assigned confidence value, at framerates above 130 Hz on commercial workstations with GPU acceleration. Pupil centre coordinates can be estimated with a median accuracy of around 1.0 pixel, and gaze estimation is accurate to within 0.5 degrees. The FCNN is able to robustly segment the pupil in a wide array of datasets that were not used for training. COMPARISON WITH EXISTING METHODS: We validate our method against gold standard eye images that were artificially rendered, as well as hand-annotated VOG data from a gold-standard clinical system (EyeSeeCam) at our institute. CONCLUSIONS: Our proposed FCNN-based pupil segmentation framework is accurate, robust and generalizes well to new VOG datasets. We provide our code and pre-trained FCNN model open-source and for free under www.github.com/pydsgz/DeepVOG.

Cloud Deployment of High-Resolution Medical Image Analysis With TOMAAT.

BACKGROUND: Deep learning has been recently applied to a multitude of computer vision and medical image analysis problems. Although recent research efforts have improved the state of the art, most of the methods cannot be easily accessed, compared or used by other researchers or clinicians. Even if developers publish their code and pre-trained models on the internet, integration in stand-alone applications and existing workflows is often not straightforward, especially for clinical research partners. In this paper, we propose an open-source framework to provide AI-enabled medical image analysis through the network. METHODS: TOMAAT provides a cloud environment for general medical image analysis, composed of three basic components: (i) an announcement service, maintaining a public registry of (ii) multiple distributed server nodes offering various medical image analysis solutions, and (iii) client software offering simple interfaces for users. Deployment is realized through HTTP-based communication, along with an API and wrappers for common image manipulations during pre- and post-processing. RESULTS: We demonstrate the utility and versatility of TOMAAT on several hallmark medical image analysis tasks: segmentation, diffeomorphic deformable atlas registration, landmark localization, and workflow integration. Through TOMAAT, the high hardware demands, setup and model complexity of demonstrated approaches are transparent to users, who are provided with simple client interfaces. We present example clients in three-dimensional Slicer, in the web browser, on iOS devices and in a commercially available, certified medical image analysis suite. CONCLUSION: TOMAAT enables deployment of state-of-the-art image segmentation in the cloud, fostering interaction among deep learning researchers and medical collaborators in the clinic. Currently, a public announcement service is hosted by the authors, and several ready-to-use services are registered and enlisted at http://tomaat.cloud.

A baseline study for detection of Parkinson's disease with 3D-transcranial sonography and uni-lateral reconstruction.

INTRODUCTION: TCS is a well-established technique for diagnosis of Parkinson's disease (PD). Volumetric 3D-TCS is a promising complementary approach for objective acquisition and analysis, in particular for less experienced sonographers. This study provides baselines for Parkinson detection (sensitivity and specificity), cutoff values and inter-rater agreement in 3D-TCS. METHODS: We performed 3D-TCS in 52 subjects (healthy controls and PD) bilaterally, and reconstructed in 3D space uni-laterally. Ipsi-lateral hyperechogenicities in the substantia nigra are manually segmented slice-by-slice in the 3D volume by two raters at different experience levels. ROC threshold analysis is performed and compared on features representing 3D volume and axial cross-sections (2.5D) of hyperechogenicities. Pearson correlation and intra-class correlation coefficients were evaluated for assessment of inter-rater agreement. RESULTS: 50 subjects were included. Both raters achieved high classification accuracy with 2.5D/3D features extracted from 3D-TCS volumes (best results sensitivity/specificity/cut-off per rater: 84.6%/88.9%/25.0mm2; 77.8%/88.9%/95.9mm3). The inter-rater agreement in 3D was high (ICC(A,1) = 0.777, p < 10-3), the classification performance of both sonographers was statistically not significantly different. CONCLUSION: The study presents first baseline values for uni-lateral 3D-TCS examination, and finds no disadvantage of uni-lateral reconstructions compared to previous bi-lateral fusion. Volumetric 3D-TCS has potential for a high inter-rater agreement and accuracy in detection of PD, in particular for sonographers with less experience.

MR imaging differentiation of Fe2+ and Fe3+ based on relaxation and magnetic susceptibility properties.

PURPOSE: The aim of this study is to evaluate the MR imaging behavior of ferrous (Fe2+) and ferric (Fe3+) iron ions in order to develop a noninvasive technique to quantitatively differentiate between both forms of iron. METHODS: MRI was performed at 3 T in a phantom consisting of 21 samples with different concentrations of ferrous and ferric chloride solutions (between 0 and 10 mmol/L). A multi-echo spoiled gradient-echo pulse sequence with eight echoes was used for both T 2* and quantitative susceptibility measurements. The transverse relaxation rate, R 2* = 1/T 2*, was determined by nonlinear exponential fitting based on the mean signals in each sample. The susceptibilities, χ, of the samples were calculated after phase unwrapping and background field removal by fitting the spatial convolution of a unit dipole response to the measured internal field map. Relaxation rate changes, ΔR 2*(c Fe), and susceptibility changes, Δχ(c Fe), their linear slopes, as well as the ratios ΔR 2*(c Fe) / Δχ(c Fe) were determined for all concentrations. RESULTS: The linear slopes of the relaxation rate were (12.5 ± 0.4) s-1/(mmol/L) for Fe3+ and (0.77 ± 0.09) s-1/(mmol/L) for Fe2+ (significantly different, z test P < 0.0001). The linear slopes of the susceptibility were (0.088 ± 0.003) ppm/(mmol/L) for Fe3+ and (0.079 ± 0.006) ppm/(mmol/L) for Fe2+. The individual ratios ΔR 2*/Δχ were greater than 40 s-1/ppm for all samples with ferric solution and lower than 20 s-1/ppm for all but one of the samples with ferrous solution. CONCLUSION: Ferrous and ferric iron ions show significantly different relaxation behaviors in MRI but similar susceptibility patterns. These properties can be used to differentiate ferrous and ferric samples.