Distribution shift detection for the postmarket surveillance of medical AI algorithms: a retrospective simulation study.

Distribution shifts remain a problem for the safe application of regulated medical AI systems, and may impact their real-world performance if undetected. Postmarket shifts can occur for example if algorithms developed on data from various acquisition settings and a heterogeneous population are predominantly applied in hospitals with lower quality data acquisition or other centre-specific acquisition factors, or where some ethnicities are over-represented. Therefore, distribution shift detection could be important for monitoring AI-based medical products during postmarket surveillance. We implemented and evaluated three deep-learning based shift detection techniques (classifier-based, deep kernel, and multiple univariate kolmogorov-smirnov tests) on simulated shifts in a dataset of 130'486 retinal images. We trained a deep learning classifier for diabetic retinopathy grading. We then simulated population shifts by changing the prevalence of patients' sex, ethnicity, and co-morbidities, and example acquisition shifts by changes in image quality. We observed classification subgroup performance disparities w.r.t. image quality, patient sex, ethnicity and co-morbidity presence. The sensitivity at detecting referable diabetic retinopathy ranged from 0.50 to 0.79 for different ethnicities. This motivates the need for detecting shifts after deployment. Classifier-based tests performed best overall, with perfect detection rates for quality and co-morbidity subgroup shifts at a sample size of 1000. It was the only method to detect shifts in patient sex, but required large sample sizes ( > 3 0 ' 000 ). All methods identified easier-to-detect out-of-distribution shifts with small (≤300) sample sizes. We conclude that effective tools exist for detecting clinically relevant distribution shifts. In particular classifier-based tests can be easily implemented components in the post-market surveillance strategy of medical device manufacturers.

An evolutionarily conserved mechanism controls reversible amyloids of pyruvate kinase via pH-sensing regions.

Amyloids are known as irreversible aggregates associated with neurodegenerative diseases. However, recent evidence shows that a subset of amyloids can form reversibly and fulfill essential cellular functions. Yet, the molecular mechanisms regulating functional amyloids and distinguishing them from pathological aggregates remain unclear. Here, we investigate the conserved principles of amyloid reversibility by studying the essential metabolic enzyme pyruvate kinase (PK) in yeast and human cells. We demonstrate that yeast PK (Cdc19) and human PK (PKM2) form reversible amyloids through a pH-sensitive amyloid core. Stress-induced cytosolic acidification promotes aggregation via protonation of specific glutamate (yeast) or histidine (human) residues within the amyloid core. Mutations mimicking protonation cause constitutive PK aggregation, while non-protonatable PK mutants remain soluble even upon stress. Physiological PK aggregation is coupled to metabolic rewiring and glycolysis arrest, causing severe growth defects when misregulated. Our work thus identifies an evolutionarily conserved, potentially widespread mechanism regulating functional amyloids during stress.

INPP5A phosphatase is a synthetic lethal target in GNAQ and GNA11-mutant melanomas.

Activating mutations in GNAQ/GNA11 occur in over 90% of uveal melanomas (UMs), the most lethal melanoma subtype; however, targeting these oncogenes has proven challenging and inhibiting their downstream effectors show limited clinical efficacy. Here, we performed genome-scale CRISPR screens along with computational analyses of cancer dependency and gene expression datasets to identify the inositol-metabolizing phosphatase INPP5A as a selective dependency in GNAQ/11-mutant UM cells in vitro and in vivo. Mutant cells intrinsically produce high levels of the second messenger inositol 1,4,5 trisphosphate (IP3) that accumulate upon suppression of INPP5A, resulting in hyperactivation of IP3-receptor signaling, increased cytosolic calcium and p53-dependent apoptosis. Finally, we show that GNAQ/11-mutant UM cells and patients' tumors exhibit elevated levels of IP4, a biomarker of enhanced IP3 production; these high levels are abolished by GNAQ/11 inhibition and correlate with sensitivity to INPP5A depletion. Our findings uncover INPP5A as a synthetic lethal vulnerability and a potential therapeutic target for GNAQ/11-mutant-driven cancers.

A reference assembly for the legume cover crop hairy vetch (Vicia villosa).

Vicia villosa is an incompletely domesticated annual legume of the Fabaceae family native to Europe and Western Asia. V. villosa is widely used as a cover crop and forage due to its ability to withstand harsh winters. Here, we generated a reference-quality genome assembly (Vvill1.0) from low error-rate long-sequence reads to improve the genetic-based trait selection of this species. Our Vvill1.0 assembly includes seven scaffolds corresponding to the seven estimated linkage groups and comprising approximately 68% of the total genome size of 2.03 Gbp. This assembly is expected to be a useful resource for genetically improving this emerging cover crop species and provide useful insights into legume genomics and plant genome evolution.

Chromosome-scale assembly of the highly heterozygous genome of red clover (Trifolium pratense L.), an allogamous forage crop species.

Relative to other crops, red clover (Trifolium pratense L.) has various favorable traits making it an ideal forage crop. Conventional breeding has improved varieties, but modern genomic methods could accelerate progress and facilitate gene discovery. Existing short-read-based genome assemblies of the ∼420 megabase pair (Mbp) genome are fragmented into >135,000 contigs, with numerous order and orientation errors within scaffolds, probably associated with the plant's biology, which displays gametophytic self-incompatibility resulting in inherent high heterozygosity. Here, we present a high-quality long-read-based assembly of red clover with a more than 500-fold reduction in contigs, improved per-base quality, and increased contig N50 by three orders of magnitude. The 413.5 Mbp assembly is nearly 20% longer than the 350 Mbp short-read assembly, closer to the predicted genome size. We also present quality measures and full-length isoform RNA transcript sequences for assessing accuracy and future genome annotation. The assembly accurately represents the seven main linkage groups in an allogamous (outcrossing), highly heterozygous plant genome.

Deep Learning Techniques for Automatic MRI Cardiac Multi-Structures Segmentation and Diagnosis: Is the Problem Solved?

Delineation of the left ventricular cavity, myocardium, and right ventricle from cardiac magnetic resonance images (multi-slice 2-D cine MRI) is a common clinical task to establish diagnosis. The automation of the corresponding tasks has thus been the subject of intense research over the past decades. In this paper, we introduce the "Automatic Cardiac Diagnosis Challenge" dataset (ACDC), the largest publicly available and fully annotated dataset for the purpose of cardiac MRI (CMR) assessment. The dataset contains data from 150 multi-equipments CMRI recordings with reference measurements and classification from two medical experts. The overarching objective of this paper is to measure how far state-of-the-art deep learning methods can go at assessing CMRI, i.e., segmenting the myocardium and the two ventricles as well as classifying pathologies. In the wake of the 2017 MICCAI-ACDC challenge, we report results from deep learning methods provided by nine research groups for the segmentation task and four groups for the classification task. Results show that the best methods faithfully reproduce the expert analysis, leading to a mean value of 0.97 correlation score for the automatic extraction of clinical indices and an accuracy of 0.96 for automatic diagnosis. These results clearly open the door to highly accurate and fully automatic analysis of cardiac CMRI. We also identify scenarios for which deep learning methods are still failing. Both the dataset and detailed results are publicly available online, while the platform will remain open for new submissions.

SonoNet: Real-Time Detection and Localisation of Fetal Standard Scan Planes in Freehand Ultrasound.

Identifying and interpreting fetal standard scan planes during 2-D ultrasound mid-pregnancy examinations are highly complex tasks, which require years of training. Apart from guiding the probe to the correct location, it can be equally difficult for a non-expert to identify relevant structures within the image. Automatic image processing can provide tools to help experienced as well as inexperienced operators with these tasks. In this paper, we propose a novel method based on convolutional neural networks, which can automatically detect 13 fetal standard views in freehand 2-D ultrasound data as well as provide a localization of the fetal structures via a bounding box. An important contribution is that the network learns to localize the target anatomy using weak supervision based on image-level labels only. The network architecture is designed to operate in real-time while providing optimal output for the localization task. We present results for real-time annotation, retrospective frame retrieval from saved videos, and localization on a very large and challenging dataset consisting of images and video recordings of full clinical anomaly screenings. We found that the proposed method achieved an average F1-score of 0.798 in a realistic classification experiment modeling real-time detection, and obtained a 90.09% accuracy for retrospective frame retrieval. Moreover, an accuracy of 77.8% was achieved on the localization task.

Multi-atlas Segmentation as a Graph Labelling Problem: Application to Partially Annotated Atlas Data.

Manually annotating images for multi-atlas segmentation is an expensive and often limiting factor in reliable automated segmentation of large databases. Segmentation methods requiring only a proportion of each atlas image to be labelled could potentially reduce the workload on expert raters tasked with labelling images. However, exploiting such a database of partially labelled atlases is not possible with state-of-the-art multi-atlas segmentation methods. In this paper we revisit the problem of multi-atlas segmentation and formulate its solution in terms of graph-labelling. Our graphical approach uses a Markov Random Field (MRF) formulation of the problem and constructs a graph connecting atlases and the target image. This provides a unifying framework for label propagation. More importantly, the proposed method can be used for segmentation using only partially labelled atlases. We furthermore provide an extension to an existing continuous MRF optimisation method to solve the proposed problem formulation. We show that the proposed method, applied to hippocampal segmentation of 202 subjects from the ADNI database, remains robust and accurate even when the proportion of manually labelled slices in the atlases is reduced to 20%.

Self-Aligning Manifolds for Matching Disparate Medical Image Datasets.

Manifold alignment can be used to reduce the dimensionality of multiple medical image datasets into a single globally consistent low-dimensional space. This may be desirable in a wide variety of problems, from fusion of different imaging modalities for Alzheimer's disease classification to 4DMR reconstruction from 2D MR slices. Unfortunately, most existing manifold alignment techniques require either a set of prior correspondences or comparability between the datasets in high-dimensional space, which is often not possible. We propose a novel technique for the 'self-alignment' of manifolds (SAM) from multiple dissimilar imaging datasets without prior correspondences or inter-dataset image comparisons. We quantitatively evaluate the method on 4DMR reconstruction from realistic, synthetic sagittal 2D MR slices from 6 volunteers and real data from 4 volunteers. Additionally, we demonstrate the technique for the compounding of two free breathing 3D ultrasound views from one volunteer. The proposed method performs significantly better for 4DMR reconstruction than state-of-the-art image-based techniques.

Mediation of clathrin-dependent trafficking during cytokinesis and cell expansion by Arabidopsis stomatal cytokinesis defective proteins.

Stomatal cytokinesis defective1 (SCD1) encodes a putative Rab guanine nucleotide exchange factor that functions in membrane trafficking and is required for cytokinesis and cell expansion in Arabidopsis thaliana. Here, we show that the loss of SCD2 function disrupts cytokinesis and cell expansion and impairs fertility, phenotypes similar to those observed for scd1 mutants. Genetic and biochemical analyses showed that SCD1 function is dependent upon SCD2 and that together these proteins are required for plasma membrane internalization. Further specifying the role of these proteins in membrane trafficking, SCD1 and SCD2 proteins were found to be associated with isolated clathrin-coated vesicles and to colocalize with clathrin light chain at putative sites of endocytosis at the plasma membrane. Together, these data suggest that SCD1 and SCD2 function in clathrin-mediated membrane transport, including plasma membrane endocytosis, required for cytokinesis and cell expansion.

SCD1 is required for cytokinesis and polarized cell expansion in Arabidopsis thaliana [corrected].

In the leaf epidermis, guard mother cells undergo a stereotyped symmetric division to form the guard cells of stomata. We have identified a temperature-sensitive Arabidopsis mutant, stomatal cytokinesis-defective 1-1 (scd1-1), which affects this specialized division. At the non-permissive temperature, 22 degrees C, defective scd1-1 guard cells are binucleate, and the formation of their ventral cell walls is incomplete. Cytokinesis was also disrupted in other types of epidermal cells such as pavement cells. Further phenotypic analysis of scd1-1 indicated a role for SCD1 in seedling growth, root elongation and flower morphogenesis. More severe scd1 T-DNA insertion alleles (scd1-2 and scd1-3) markedly affect polar cell expansion, most notably in trichomes and root hairs. SCD1 is a unique gene in Arabidopsis that encodes a protein related to animal proteins that regulate intracellular protein transport and/or mitogen-activated protein kinase signaling pathways. Consistent with a role for SCD1 in membrane trafficking, secretory vesicles were found to accumulate in cytokinesis-defective scd1 cells. In addition the scd1 mutant phenotype was enhanced by low doses of inhibitors of cell plate consolidation and vesicle secretion. We propose that SCD1 functions in polarized vesicle trafficking during plant cytokinesis and cell expansion.