A unified framework for estimating country-specific cumulative incidence for 18 diseases stratified by polygenic risk.

Polygenic scores (PGSs) offer the ability to predict genetic risk for complex diseases across the life course; a key benefit over short-term prediction models. To produce risk estimates relevant to clinical and public health decision-making, it is important to account for varying effects due to age and sex. Here, we develop a novel framework to estimate country-, age-, and sex-specific estimates of cumulative incidence stratified by PGS for 18 high-burden diseases. We integrate PGS associations from seven studies in four countries (N = 1,197,129) with disease incidences from the Global Burden of Disease. PGS has a significant sex-specific effect for asthma, hip osteoarthritis, gout, coronary heart disease and type 2 diabetes (T2D), with all but T2D exhibiting a larger effect in men. PGS has a larger effect in younger individuals for 13 diseases, with effects decreasing linearly with age. We show for breast cancer that, relative to individuals in the bottom 20% of polygenic risk, the top 5% attain an absolute risk for screening eligibility 16.3 years earlier. Our framework increases the generalizability of results from biobank studies and the accuracy of absolute risk estimates by appropriately accounting for age- and sex-specific PGS effects. Our results highlight the potential of PGS as a screening tool which may assist in the early prevention of common diseases.

Evaluation of polygenic scoring methods in five biobanks shows larger variation between biobanks than methods and finds benefits of ensemble learning.

Methods of estimating polygenic scores (PGSs) from genome-wide association studies are increasingly utilized. However, independent method evaluation is lacking, and method comparisons are often limited. Here, we evaluate polygenic scores derived via seven methods in five biobank studies (totaling about 1.2 million participants) across 16 diseases and quantitative traits, building on a reference-standardized framework. We conducted meta-analyses to quantify the effects of method choice, hyperparameter tuning, method ensembling, and the target biobank on PGS performance. We found that no single method consistently outperformed all others. PGS effect sizes were more variable between biobanks than between methods within biobanks when methods were well tuned. Differences between methods were largest for the two investigated autoimmune diseases, seropositive rheumatoid arthritis and type 1 diabetes. For most methods, cross-validation was more reliable for tuning hyperparameters than automatic tuning (without the use of target data). For a given target phenotype, elastic net models combining PGS across methods (ensemble PGS) tuned in the UK Biobank provided consistent, high, and cross-biobank transferable performance, increasing PGS effect sizes (ß coefficients) by a median of 5.0% relative to LDpred2 and MegaPRS (the two best-performing single methods when tuned with cross-validation). Our interactively browsable online-results and open-source workflow prspipe provide a rich resource and reference for the analysis of polygenic scoring methods across biobanks.

Simulating rigid head motion artifacts on brain magnitude MRI data-Outcome on image quality and segmentation of the cerebral cortex.

Magnetic Resonance Imaging (MRI) datasets from epidemiological studies often show a lower prevalence of motion artifacts than what is encountered in clinical practice. These artifacts can be unevenly distributed between subject groups and studies which introduces a bias that needs addressing when augmenting data for machine learning purposes. Since unreconstructed multi-channel k-space data is typically not available for population-based MRI datasets, motion simulations must be performed using signal magnitude data. There is thus a need to systematically evaluate how realistic such magnitude-based simulations are. We performed magnitude-based motion simulations on a dataset (MR-ART) from 148 subjects in which real motion-corrupted reference data was also available. The similarity of real and simulated motion was assessed by using image quality metrics (IQMs) including Coefficient of Joint Variation (CJV), Signal-to-Noise-Ratio (SNR), and Contrast-to-Noise-Ratio (CNR). An additional comparison was made by investigating the decrease in the Dice-Sørensen Coefficient (DSC) of automated segmentations with increasing motion severity. Segmentation of the cerebral cortex was performed with 6 freely available tools: FreeSurfer, BrainSuite, ANTs, SAMSEG, FastSurfer, and SynthSeg+. To better mimic the real subject motion, the original motion simulation within an existing data augmentation framework (TorchIO), was modified. This allowed a non-random motion paradigm and phase encoding direction. The mean difference in CJV/SNR/CNR between the real motion-corrupted images and our modified simulations (0.004±0.054/-0.7±1.8/-0.09±0.55) was lower than that of the original simulations (0.015±0.061/0.2±2.0/-0.29±0.62). Further, the mean difference in the DSC between the real motion-corrupted images was lower for our modified simulations (0.03±0.06) compared to the original simulations (-0.15±0.09). SynthSeg+ showed the highest robustness towards all forms of motion, real and simulated. In conclusion, reasonably realistic synthetic motion artifacts can be induced on a large-scale when only magnitude MR images are available to obtain unbiased data sets for the training of machine learning based models.

A trustworthy AI reality-check: the lack of transparency of artificial intelligence products in healthcare.

Trustworthy medical AI requires transparency about the development and testing of underlying algorithms to identify biases and communicate potential risks of harm. Abundant guidance exists on how to achieve transparency for medical AI products, but it is unclear whether publicly available information adequately informs about their risks. To assess this, we retrieved public documentation on the 14 available CE-certified AI-based radiology products of the II b risk category in the EU from vendor websites, scientific publications, and the European EUDAMED database. Using a self-designed survey, we reported on their development, validation, ethical considerations, and deployment caveats, according to trustworthy AI guidelines. We scored each question with either 0, 0.5, or 1, to rate if the required information was "unavailable", "partially available," or "fully available." The transparency of each product was calculated relative to all 55 questions. Transparency scores ranged from 6.4% to 60.9%, with a median of 29.1%. Major transparency gaps included missing documentation on training data, ethical considerations, and limitations for deployment. Ethical aspects like consent, safety monitoring, and GDPR-compliance were rarely documented. Furthermore, deployment caveats for different demographics and medical settings were scarce. In conclusion, public documentation of authorized medical AI products in Europe lacks sufficient public transparency to inform about safety and risks. We call on lawmakers and regulators to establish legally mandated requirements for public and substantive transparency to fulfill the promise of trustworthy AI for health.

Dynamic parametric MRI and deep learning: Unveiling renal pathophysiology through accurate kidney size quantification.

Renal pathologies often manifest as alterations in kidney size, providing a valuable avenue for employing dynamic parametric MRI as a means to derive kidney size measurements for the diagnosis, treatment, and monitoring of renal disease. Furthermore, this approach holds significant potential in supporting MRI data-driven preclinical investigations into the intricate mechanisms underlying renal pathophysiology. The integration of deep learning algorithms is crucial in achieving rapid and precise segmentation of the kidney from temporally resolved parametric MRI, facilitating the use of kidney size as a meaningful (pre)clinical biomarker for renal disease. To explore this potential, we employed dynamic parametric T2 mapping of the kidney in rats in conjunction with a custom-tailored deep dilated U-Net (DDU-Net) architecture. The architecture was trained, validated, and tested on manually segmented ground truth kidney data, with benchmarking against an analytical segmentation model and a self-configuring no new U-Net. Subsequently, we applied our approach to in vivo longitudinal MRI data, incorporating interventions that emulate clinically relevant scenarios in rats. Our approach achieved high performance metrics, including a Dice coefficient of 0.98, coefficient of determination of 0.92, and a mean absolute percentage error of 1.1% compared with ground truth. The DDU-Net enabled automated and accurate quantification of acute changes in kidney size, such as aortic occlusion (-8% ± 1%), venous occlusion (5% ± 1%), furosemide administration (2% ± 1%), hypoxemia (-2% ± 1%), and contrast agent-induced acute kidney injury (11% ± 1%). This approach can potentially be instrumental for the development of dynamic parametric MRI-based tools for kidney disorders, offering unparalleled insights into renal pathophysiology.

HAPNEST: efficient, large-scale generation and evaluation of synthetic datasets for genotypes and phenotypes.

MOTIVATION: Existing methods for simulating synthetic genotype and phenotype datasets have limited scalability, constraining their usability for large-scale analyses. Moreover, a systematic approach for evaluating synthetic data quality and a benchmark synthetic dataset for developing and evaluating methods for polygenic risk scores are lacking. RESULTS: We present HAPNEST, a novel approach for efficiently generating diverse individual-level genotypic and phenotypic data. In comparison to alternative methods, HAPNEST shows faster computational speed and a lower degree of relatedness with reference panels, while generating datasets that preserve key statistical properties of real data. These desirable synthetic data properties enabled us to generate 6.8 million common variants and nine phenotypes with varying degrees of heritability and polygenicity across 1 million individuals. We demonstrate how HAPNEST can facilitate biobank-scale analyses through the comparison of seven methods to generate polygenic risk scoring across multiple ancestry groups and different genetic architectures. AVAILABILITY AND IMPLEMENTATION: A synthetic dataset of 1 008 000 individuals and nine traits for 6.8 million common variants is available at https://www.ebi.ac.uk/biostudies/studies/S-BSST936. The HAPNEST software for generating synthetic datasets is available as Docker/Singularity containers and open source Julia and C code at https://github.com/intervene-EU-H2020/synthetic_data.

Improved prediction of glioma-related aphasia by diffusion MRI metrics, machine learning, and automated fiber bundle segmentation.

White matter impairments caused by gliomas can lead to functional disorders. In this study, we predicted aphasia in patients with gliomas infiltrating the language network using machine learning methods. We included 78 patients with left-hemispheric perisylvian gliomas. Aphasia was graded preoperatively using the Aachen aphasia test (AAT). Subsequently, we created bundle segmentations based on automatically generated tract orientation mappings using TractSeg. To prepare the input for the support vector machine (SVM), we first preselected aphasia-related fiber bundles based on the associations between relative tract volumes and AAT subtests. In addition, diffusion magnetic resonance imaging (dMRI)-based metrics [axial diffusivity (AD), apparent diffusion coefficient (ADC), fractional anisotropy (FA), and radial diffusivity (RD)] were extracted within the fiber bundles' masks with their mean, standard deviation, kurtosis, and skewness values. Our model consisted of random forest-based feature selection followed by an SVM. The best model performance achieved 81% accuracy (specificity = 85%, sensitivity = 73%, and AUC = 85%) using dMRI-based features, demographics, tumor WHO grade, tumor location, and relative tract volumes. The most effective features resulted from the arcuate fasciculus (AF), middle longitudinal fasciculus (MLF), and inferior fronto-occipital fasciculus (IFOF). The most effective dMRI-based metrics were FA, ADC, and AD. We achieved a prediction of aphasia using dMRI-based features and demonstrated that AF, IFOF, and MLF were the most important fiber bundles for predicting aphasia in this cohort.

A deep-learning algorithm to classify skin lesions from mpox virus infection.

Undetected infection and delayed isolation of infected individuals are key factors driving the monkeypox virus (now termed mpox virus or MPXV) outbreak. To enable earlier detection of MPXV infection, we developed an image-based deep convolutional neural network (named MPXV-CNN) for the identification of the characteristic skin lesions caused by MPXV. We assembled a dataset of 139,198 skin lesion images, split into training/validation and testing cohorts, comprising non-MPXV images (n = 138,522) from eight dermatological repositories and MPXV images (n = 676) from the scientific literature, news articles, social media and a prospective cohort of the Stanford University Medical Center (n = 63 images from 12 patients, all male). In the validation and testing cohorts, the sensitivity of the MPXV-CNN was 0.83 and 0.91, the specificity was 0.965 and 0.898 and the area under the curve was 0.967 and 0.966, respectively. In the prospective cohort, the sensitivity was 0.89. The classification performance of the MPXV-CNN was robust across various skin tones and body regions. To facilitate the usage of the algorithm, we developed a web-based app by which the MPXV-CNN can be accessed for patient guidance. The capability of the MPXV-CNN for identifying MPXV lesions has the potential to aid in MPXV outbreak mitigation.

First in vivo fluorine-19 magnetic resonance imaging of the multiple sclerosis drug siponimod.

Theranostic imaging methods could greatly enhance our understanding of the distribution of CNS-acting drugs in individual patients. Fluorine-19 magnetic resonance imaging (19F MRI) offers the opportunity to localize and quantify fluorinated drugs non-invasively, without modifications and without the application of ionizing or other harmful radiation. Here we investigated siponimod, a sphingosine 1-phosphate (S1P) receptor antagonist indicated for secondary progressive multiple sclerosis (SPMS), to determine the feasibility of in vivo 19F MR imaging of a disease modifying drug. Methods: The 19F MR properties of siponimod were characterized using spectroscopic techniques. Four MRI methods were investigated to determine which was the most sensitive for 19F MR imaging of siponimod under biological conditions. We subsequently administered siponimod orally to 6 mice and acquired 19F MR spectra and images in vivo directly after administration, and in ex vivo tissues. Results: The 19F transverse relaxation time of siponimod was 381 ms when dissolved in dimethyl sulfoxide, and substantially reduced to 5 ms when combined with serum, and to 20 ms in ex vivo liver tissue. Ultrashort echo time (UTE) imaging was determined to be the most sensitive MRI technique for imaging siponimod in a biological context and was used to map the drug in vivo in the stomach and liver. Ex vivo images in the liver and brain showed an inhomogeneous distribution of siponimod in both organs. In the brain, siponimod accumulated predominantly in the cerebrum but not the cerebellum. No secondary 19F signals were detected from metabolites. From a translational perspective, we found that acquisitions done on a 3.0 T clinical MR scanner were 2.75 times more sensitive than acquisitions performed on a preclinical 9.4 T MR setup when taking changes in brain size across species into consideration and using equivalent relative spatial resolution. Conclusion: Siponimod can be imaged non-invasively using 19F UTE MRI in the form administered to MS patients, without modification. This study lays the groundwork for more extensive preclinical and clinical investigations. With the necessary technical development, 19F MRI has the potential to become a powerful theranostic tool for studying the time-course and distribution of CNS-acting drugs within the brain, especially during pathology.

AntiSplodge: a neural-network-based RNA-profile deconvolution pipeline designed for spatial transcriptomics.

With the current surge of spatial transcriptomics (ST) studies, researchers are exploring the deep interactive cell-play directly in tissues, in situ. However, with the current technologies, measurements consist of mRNA transcript profiles of mixed origin. Recently, applications have been proposed to tackle the deconvolution process, to gain knowledge about which cell types (SC) are found within. This is usually done by incorporating metrics from single-cell (SC) RNA, from similar tissues. Yet, most existing tools are cumbersome, and we found them hard to integrate and properly utilize. Therefore, we present AntiSplodge, a simple feed-forward neural-network-based pipeline designed to effective deconvolute ST profiles by utilizing synthetic ST profiles derived from real-life SC datasets. AntiSplodge is designed to be easy, fast and intuitive while still being lightweight. To demonstrate AntiSplodge, we deconvolute the human heart and verify correctness across time points. We further deconvolute the mouse brain, where spot patterns correctly follow that of the underlying tissue. In particular, for the hippocampus from where the cells originate. Furthermore, AntiSplodge demonstrates top of the line performance when compared to current state-of-the-art tools. Software availability: https://github.com/HealthML/AntiSplodge/.

Piloting a Survey-Based Assessment of Transparency and Trustworthiness with Three Medical AI Tools.

Artificial intelligence (AI) offers the potential to support healthcare delivery, but poorly trained or validated algorithms bear risks of harm. Ethical guidelines stated transparency about model development and validation as a requirement for trustworthy AI. Abundant guidance exists to provide transparency through reporting, but poorly reported medical AI tools are common. To close this transparency gap, we developed and piloted a framework to quantify the transparency of medical AI tools with three use cases. Our framework comprises a survey to report on the intended use, training and validation data and processes, ethical considerations, and deployment recommendations. The transparency of each response was scored with either 0, 0.5, or 1 to reflect if the requested information was not, partially, or fully provided. Additionally, we assessed on an analogous three-point scale if the provided responses fulfilled the transparency requirement for a set of trustworthiness criteria from ethical guidelines. The degree of transparency and trustworthiness was calculated on a scale from 0% to 100%. Our assessment of three medical AI use cases pin-pointed reporting gaps and resulted in transparency scores of 67% for two use cases and one with 59%. We report anecdotal evidence that business constraints and limited information from external datasets were major obstacles to providing transparency for the three use cases. The observed transparency gaps also lowered the degree of trustworthiness, indicating compliance gaps with ethical guidelines. All three pilot use cases faced challenges to provide transparency about medical AI tools, but more studies are needed to investigate those in the wider medical AI sector. Applying this framework for an external assessment of transparency may be infeasible if business constraints prevent the disclosure of information. New strategies may be necessary to enable audits of medical AI tools while preserving business secrets.

Identifying interpretable gene-biomarker associations with functionally informed kernel-based tests in 190,000 exomes.

Here we present an exome-wide rare genetic variant association study for 30 blood biomarkers in 191,971 individuals in the UK Biobank. We compare gene-based association tests for separate functional variant categories to increase interpretability and identify 193 significant gene-biomarker associations. Genes associated with biomarkers were ~ 4.5-fold enriched for conferring Mendelian disorders. In addition to performing weighted gene-based variant collapsing tests, we design and apply variant-category-specific kernel-based tests that integrate quantitative functional variant effect predictions for missense variants, splicing and the binding of RNA-binding proteins. For these tests, we present a computationally efficient combination of the likelihood-ratio and score tests that found 36% more associations than the score test alone while also controlling the type-1 error. Kernel-based tests identified 13% more associations than their gene-based collapsing counterparts and had advantages in the presence of gain of function missense variants. We introduce local collapsing by amino acid position for missense variants and use it to interpret associations and identify potential novel gain of function variants in PIEZO1. Our results show the benefits of investigating different functional mechanisms when performing rare-variant association tests, and demonstrate pervasive rare-variant contribution to biomarker variability.

Pathogenic variants damage cell composition and single cell transcription in cardiomyopathies.

Pathogenic variants in genes that cause dilated cardiomyopathy (DCM) and arrhythmogenic cardiomyopathy (ACM) convey high risks for the development of heart failure through unknown mechanisms. Using single-nucleus RNA sequencing, we characterized the transcriptome of 880,000 nuclei from 18 control and 61 failing, nonischemic human hearts with pathogenic variants in DCM and ACM genes or idiopathic disease. We performed genotype-stratified analyses of the ventricular cell lineages and transcriptional states. The resultant DCM and ACM ventricular cell atlas demonstrated distinct right and left ventricular responses, highlighting genotype-associated pathways, intercellular interactions, and differential gene expression at single-cell resolution. Together, these data illuminate both shared and distinct cellular and molecular architectures of human heart failure and suggest candidate therapeutic targets.

Identification of novel genes whose expression in adipose tissue affects body fat mass and distribution: an RNA-Seq and Mendelian Randomization study.

Many studies have shown that abdominal adiposity is more strongly related to health risks than peripheral adiposity. However, the underlying pathways are still poorly understood. In this cross-sectional study using data from RNA-sequencing experiments and whole-body MRI scans of 200 participants in the EPIC-Potsdam cohort, our aim was to identify novel genes whose gene expression in subcutaneous adipose tissue has an effect on body fat mass (BFM) and body fat distribution (BFD). The analysis identified 625 genes associated with adiposity, of which 531 encode a known protein and 487 are novel candidate genes for obesity. Enrichment analyses indicated that BFM-associated genes were characterized by their higher than expected involvement in cellular, regulatory and immune system processes, and BFD-associated genes by their involvement in cellular, metabolic, and regulatory processes. Mendelian Randomization analyses suggested that the gene expression of 69 genes was causally related to BFM and BFD. Six genes were replicated in UK Biobank. In this study, we identified novel genes for BFM and BFD that are BFM- and BFD-specific, involved in different molecular processes, and whose up-/downregulated gene expression may causally contribute to obesity.

Machine learning-based prediction of motor status in glioma patients using diffusion MRI metrics along the corticospinal tract.

Along tract statistics enables white matter characterization using various diffusion MRI metrics. These diffusion models reveal detailed insights into white matter microstructural changes with development, pathology and function. Here, we aim at assessing the clinical utility of diffusion MRI metrics along the corticospinal tract, investigating whether motor glioma patients can be classified with respect to their motor status. We retrospectively included 116 brain tumour patients suffering from either left or right supratentorial, unilateral World Health Organization Grades II, III and IV gliomas with a mean age of 53.51 ± 16.32 years. Around 37% of patients presented with preoperative motor function deficits according to the Medical Research Council scale. At group level comparison, the highest non-overlapping diffusion MRI differences were detected in the superior portion of the tracts' profiles. Fractional anisotropy and fibre density decrease, apparent diffusion coefficient axial diffusivity and radial diffusivity increase. To predict motor deficits, we developed a method based on a support vector machine using histogram-based features of diffusion MRI tract profiles (e.g. mean, standard deviation, kurtosis and skewness), following a recursive feature elimination method. Our model achieved high performance (74% sensitivity, 75% specificity, 74% overall accuracy and 77% area under the curve). We found that apparent diffusion coefficient, fractional anisotropy and radial diffusivity contributed more than other features to the model. Incorporating the patient demographics and clinical features such as age, tumour World Health Organization grade, tumour location, gender and resting motor threshold did not affect the model's performance, revealing that these features were not as effective as microstructural measures. These results shed light on the potential patterns of tumour-related microstructural white matter changes in the prediction of functional deficits.

Self-Supervised Learning Methods for Label-Efficient Dental Caries Classification.

High annotation costs are a substantial bottleneck in applying deep learning architectures to clinically relevant use cases, substantiating the need for algorithms to learn from unlabeled data. In this work, we propose employing self-supervised methods. To that end, we trained with three self-supervised algorithms on a large corpus of unlabeled dental images, which contained 38K bitewing radiographs (BWRs). We then applied the learned neural network representations on tooth-level dental caries classification, for which we utilized labels extracted from electronic health records (EHRs). Finally, a holdout test-set was established, which consisted of 343 BWRs and was annotated by three dental professionals and approved by a senior dentist. This test-set was used to evaluate the fine-tuned caries classification models. Our experimental results demonstrate the obtained gains by pretraining models using self-supervised algorithms. These include improved caries classification performance (6 p.p. increase in sensitivity) and, most importantly, improved label-efficiency. In other words, the resulting models can be fine-tuned using few labels (annotations). Our results show that using as few as 18 annotations can produce ≥45% sensitivity, which is comparable to human-level diagnostic performance. This study shows that self-supervision can provide gains in medical image analysis, particularly when obtaining labels is costly and expensive.

transferGWAS: GWAS of images using deep transfer learning.

MOTIVATION: Medical images can provide rich information about diseases and their biology. However, investigating their association with genetic variation requires non-standard methods. We propose transferGWAS, a novel approach to perform genome-wide association studies directly on full medical images. First, we learn semantically meaningful representations of the images based on a transfer learning task, during which a deep neural network is trained on independent but similar data. Then, we perform genetic association tests with these representations. RESULTS: We validate the type I error rates and power of transferGWAS in simulation studies of synthetic images. Then we apply transferGWAS in a genome-wide association study of retinal fundus images from the UK Biobank. This first-of-a-kind GWAS of full imaging data yielded 60 genomic regions associated with retinal fundus images, of which 7 are novel candidate loci for eye-related traits and diseases. AVAILABILITY AND IMPLEMENTATION: Our method is implemented in Python and available at https://github.com/mkirchler/transferGWAS/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Predicting the SARS-CoV-2 effective reproduction number using bulk contact data from mobile phones.

Over the last months, cases of SARS-CoV-2 surged repeatedly in many countries but could often be controlled with nonpharmaceutical interventions including social distancing. We analyzed deidentified Global Positioning System (GPS) tracking data from 1.15 to 1.4 million cell phones in Germany per day between March and November 2020 to identify encounters between individuals and statistically evaluate contact behavior. Using graph sampling theory, we estimated the contact index (CX), a metric for number and heterogeneity of contacts. We found that CX, and not the total number of contacts, is an accurate predictor for the effective reproduction number R derived from case numbers. A high correlation between CX and R recorded more than 2 wk later allows assessment of social behavior well before changes in case numbers become detectable. By construction, the CX quantifies the role of superspreading and permits assigning risks to specific contact behavior. We provide a critical CX value beyond which R is expected to rise above 1 and propose to use that value to leverage the social-distancing interventions for the coming months.