Bayesian Segmentation Research Articles

Histology-guided MRI segmentation of brainstem nuclei critical to consciousness.

While substantial progress has been made in mapping the connectivity of cortical networks responsible for conscious awareness, neuroimaging analysis of subcortical arousal networks that modulate arousal (i.e., wakefulness) has been limited by a lack of a robust segmentation procedures for brainstem arousal nuclei. Automated segmentation of brainstem arousal nuclei is an essential step toward elucidating the physiology of arousal in human consciousness and the pathophysiology of disorders of consciousness. We created a probabilistic atlas of brainstem arousal nuclei built on diffusion MRI scans of five ex vivo human brain specimens scanned at 750 μm isotropic resolution. Labels of arousal nuclei used to generate the probabilistic atlas were manually annotated with reference to nucleus-specific immunostaining in two of the five brain specimens. We then developed a Bayesian segmentation algorithm that utilizes the probabilistic atlas as a generative model and automatically identifies brainstem arousal nuclei in a resolution- and contrast-agnostic manner. The segmentation method displayed high accuracy in both healthy and lesioned in vivo T1 MRI scans and high test-retest reliability across both T1 and T2 MRI contrasts. Finally, we show that the segmentation algorithm can detect volumetric changes and differences in magnetic susceptibility within brainstem arousal nuclei in Alzheimer's disease and traumatic coma, respectively. We release the probabilistic atlas and Bayesian segmentation tool in FreeSurfer to advance the study of human consciousness and its disorders.

Hypothalamic Subunit Volumes in Schizophrenia and Bipolar Spectrum Disorders.

The hypothalamus is central to many hormonal and autonomous nervous system pathways. Emerging evidence indicates that these pathways may be disrupted in schizophrenia and bipolar disorder. Yet, few studies have examined the volumes of hypothalamic subunits in these patient groups. We compared hypothalamic subunit volumes in individuals with psychotic disorders to healthy controls. We included 344 patients with schizophrenia spectrum disorders (SCZ), 340 patients with bipolar disorders (BPD), and 684 age- and-sex-matched healthy controls (CTR). Total hypothalamus and five hypothalamic subunit volumes were extracted from T1-weighted magnetic resonance imaging (MRI) using an automated Bayesian segmentation method. Regression models, corrected for age, age2, sex, and segmentation-based intracranial volume (sbTIV), were used to examine diagnostic group differences, interactions with sex, and associations with clinical symptoms, antipsychotic medication, antidepressants and mood stabilizers. SCZ had larger volumes in the left inferior tubular subunit and smaller right anterior-inferior, right anterior-superior, and right posterior hypothalamic subunits compared to CTR. BPD did not differ significantly from CTR for any hypothalamic subunit volume, however, there was a significant sex-by-diagnosis interaction. Analyses stratified by sex showed smaller right hypothalamus and right posterior subunit volumes in male patients, but not female patients, relative to same-sex controls. There was a significant association between BPD currently taking antipsychotic medication and the left inferior tubular subunits volumes. Our results show regional-specific alterations in hypothalamus subunit volumes in individuals with SCZ, with relevance to HPA-axis dysregulation, circadian rhythm disruption, and cognition impairment.

End-to-End Bayesian Segmentation and Similarity Assessment of Performed Music Tempo and Dynamics without Score Information

Segmenting continuous sensory input into coherent segments and subsegments is an important part of perception. Music is no exception. By shaping the acoustic properties of music during performance, musicians can strongly influence the perceived segmentation. Two main techniques musicians employ are the modulation of tempo and dynamics. Such variations carry important information for segmentation and lend themselves well to numerical analysis methods. In this article, based on tempo or loudness modulations alone, we propose a novel end-to-end Bayesian framework using dynamic programming to retrieve a musician's expressed segmentation. The method computes the credence of all possible segmentations of the recorded performance. The output is summarized in two forms: as a beat-by-beat profile revealing the posterior credence of plausible boundaries, and as expanded credence segment maps, a novel representation that converts readily to a segmentation lattice but retains information about the posterior uncertainty on the exact position of segments’ endpoints. To compare any two segmentation profiles, we introduce a method based on unbalanced optimal transport. Experimental results on the MazurkaBL dataset show that despite the drastic dimension reduction from the input data, the segmentation recovery is sufficient for deriving musical insights from comparative examination of recorded performances. This Bayesian segmentation method thus offers an alternative to binary boundary detection and finds multiple hypotheses fitting information from recorded music performances.

BayeSeg: Bayesian modeling for medical image segmentation with interpretable generalizability.

Due to the cross-domain distribution shift aroused from diverse medical imaging systems, many deep learning segmentation methods fail to perform well on unseen data, which limits their real-world applicability. Recent works have shown the benefits of extracting domain-invariant representations on domain generalization. However, the interpretability of domain-invariant features remains a great challenge. To address this problem, we propose an interpretable Bayesian framework (BayeSeg) through Bayesian modeling of image and label statistics to enhance model generalizability for medical image segmentation. Specifically, we first decompose an image into a spatial-correlated variable and a spatial-variant variable, assigning hierarchical Bayesian priors to explicitly force them to model the domain-stable shape and domain-specific appearance information respectively. Then, we model the segmentation as a locally smooth variable only related to the shape. Finally, we develop a variational Bayesian framework to infer the posterior distributions of these explainable variables. The framework is implemented with neural networks, and thus is referred to as deep Bayesian segmentation. Quantitative and qualitative experimental results on prostate segmentation and cardiac segmentation tasks have shown the effectiveness of our proposed method. Moreover, we investigated the interpretability of BayeSeg by explaining the posteriors and analyzed certain factors that affect the generalization ability through further ablation studies. Our code is released via https://zmiclab.github.io/projects.html.

Alterations of brainstem volume in patients with first-episode and recurrent major depressive disorder

BackgroundMajor depressive disorder (MDD) is a prevalent mental health condition characterized by recurrent episodes in a substantial proportion of patients. The number of previous episodes is one of the most crucial predictors of depression recurrence. However, the underlying neural mechanisms remain unclear. To date, there have been limited neuroimaging studies investigating morphological changes of the brainstem in patients with first-episode MDD (FMDD) and recurrent MDD (RMDD). This study aimed to examine volumetric changes of individual brainstem regions in relation to the number of previous episodes and disease duration.MethodA total of 111 individuals including 36 FMDD, 25 RMDD, and 50 healthy controls (HCs) underwent T1-weighted structural magnetic resonance imaging scans. A Bayesian segmentation algorithm was used to analyze the volume of each brainstem region, including the medulla oblongata, pons, midbrain, and superior cerebellar peduncle (SCP), as well as the whole brainstem volume. Analyses of variance (ANOVA) were performed to obtain brain regions with significant differences among three groups and then post hoc tests were calculated for inter-group comparisons. Partial correlation analyses were further conducted to identify associations between regional volumes and clinical features.ResultsThe ANOVA revealed significant brainstem volumetric differences among three groups in the pons, midbrain, SCP, and the whole brainstem (F = 3.996 ~ 5.886, adjusted p = 0.015 ~ 0.028). As compared with HCs, both groups of MDD patients showed decreased volumes in the pons as well as the entire brainstem (p = 0.002 ~ 0.034), however, only the FMDD group demonstrated a significantly reduced volume in the midbrain (p = 0.003). Specifically, the RMDD group exhibited significantly decreased SCP volume when comparing to both FMDD (p = 0.021) group and HCs (p = 0.008). Correlation analyses revealed that the SCP volumes were negatively associated with the number of depressive episodes (r=-0.36, p < 0.01) and illness duration (r=-0.28, p = 0.035) in patients with MDD.ConclusionThe present findings provided evidence of decreased brainstem volume involving in the pathophysiology of MDD, particularly, volumetric reduction in the SCP might represent a neurobiological marker for RMDD. Further research is needed to confirm our observations and deepen our understanding of the neural mechanisms underlying depression recurrence.

Thalamic nuclei volumes in schizophrenia and bipolar spectrum disorders - Associations with diagnosis and clinical characteristics.

The thalamus is central to brain functions ranging from primary sensory processing to higher-order cognition. Structural deficits in thalamic association nuclei such as the pulvinar and mediodorsal nuclei have previously been reported in schizophrenia. However, the specificity with regards to clinical presentation, and whether or not bipolar disorder (BD) is associated with similar alterations is unclear. We investigated thalamic nuclei volumes in 334 patients with schizophrenia spectrum disorders (SSD) (median age 29years, 59% male), 322 patients with BD (30years, 40% male), and 826 healthy controls (HC) (34years, 54% male). Volumes of 25 thalamic nuclei were extracted from T1-weighted magnetic resonance imaging using an automated Bayesian segmentation method and compared between groups. Furthermore, we explored associations with clinical characteristics across diagnostic groups, including psychotic and mood symptoms and medication use, as well as diagnostic subtype in BD. Significantly smaller volumes were found in the mediodorsal, pulvinar, and lateral and medial geniculate thalamic nuclei in SSD. Similarly, smaller volumes were found in BD in the same four regions, but mediodorsal nucleus volume alterations were limited to its lateral part and pulvinar alterations to its anterior region. Smaller volumes in BD compared to HC were seen only in BD type I, not BD type II. Across diagnoses, having more negative symptoms was associated with smaller pulvinar volumes. Structural alterations were found in both SSD and BD, mainly in the thalamic association nuclei. Structural deficits in the pulvinar may be of relevance for negative symptoms.

SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining.

Despite advances in data augmentation and transfer learning, convolutional neural networks (CNNs) difficultly generalise to unseen domains. When segmenting brain scans, CNNs are highly sensitive to changes in resolution and contrast: even within the same MRI modality, performance can decrease across datasets. Here we introduce SynthSeg, the first segmentation CNN robust against changes in contrast and resolution. SynthSeg is trained with synthetic data sampled from a generative model conditioned on segmentations. Crucially, we adopt a domain randomisation strategy where we fully randomise the contrast and resolution of the synthetic training data. Consequently, SynthSeg can segment real scans from a wide range of target domains without retraining or fine-tuning, which enables straightforward analysis of huge amounts of heterogeneous clinical data. Because SynthSeg only requires segmentations to be trained (no images), it can learn from labels obtained by automated methods on diverse populations (e.g., ageing and diseased), thus achieving robustness to a wide range of morphological variability. We demonstrate SynthSeg on 5,000 scans of six modalities (including CT) and ten resolutions, where it exhibits unparallelled generalisation compared with supervised CNNs, state-of-the-art domain adaptation, and Bayesian segmentation. Finally, we demonstrate the generalisability of SynthSeg by applying it to cardiac MRI and CTscans.

Accurate Bayesian segmentation of thalamic nuclei using diffusion MRI and an improved histological atlas

The human thalamus is a highly connected brain structure, which is key for the control of numerous functions and is involved in several neurological disorders. Recently, neuroimaging studies have increasingly focused on the volume and connectivity of the specific nuclei comprising this structure, rather than looking at the thalamus as a whole. However, accurate identification of cytoarchitectonically designed histological nuclei on standard in vivo structural MRI is hampered by the lack of image contrast that can be used to distinguish nuclei from each other and from surrounding white matter tracts. While diffusion MRI may offer such contrast, it has lower resolution and lacks some boundaries visible in structural imaging. In this work, we present a Bayesian segmentation algorithm for the thalamus. This algorithm combines prior information from a probabilistic atlas with likelihood models for both structural and diffusion MRI, allowing segmentation of 25 thalamic labels per hemisphere informed by both modalities. We present an improved probabilistic atlas, incorporating thalamic nuclei identified from histology and 45 white matter tracts surrounding the thalamus identified in ultra-high gradient strength diffusion imaging. We present a family of likelihood models for diffusion tensor imaging, ensuring compatibility with the vast majority of neuroimaging datasets that include diffusion MRI data. The use of these diffusion likelihood models greatly improves identification of nuclear groups versus segmentation based solely on structural MRI. Dice comparison of 5 manually identifiable groups of nuclei to ground truth segmentations show improvements of up to 10 percentage points. Additionally, our chosen model shows a high degree of reliability, with median test-retest Dice scores above 0.85 for four out of five nuclei groups, whilst also offering improved detection of differential thalamic involvement in Alzheimer’s disease (AUROC 81.98%). The probabilistic atlas and segmentation tool will be made publicly available as part of the neuroimaging package FreeSurfer (https://freesurfer.net/fswiki/ThalamicNucleiDTI).

Evaluation of Semiautomatic and Deep Learning-Based Fully Automatic Segmentation Methods on [18F]FDG PET/CT Images from Patients with Lymphoma: Influence on Tumor Characterization.

The objective is to assess the performance of seven semiautomatic and two fully automatic segmentation methods on [18F]FDG PET/CT lymphoma images and evaluate their influence on tumor quantification. All lymphoma lesions identified in 65 whole-body [18F]FDG PET/CT staging images were segmented by two experienced observers using manual and semiautomatic methods. Semiautomatic segmentation using absolute and relative thresholds, k-means and Bayesian clustering, and a self-adaptive configuration (SAC) of k-means and Bayesian was applied. Three state-of-the-art deep learning-based segmentations methods using a 3D U-Net architecture were also applied. One was semiautomatic and two were fully automatic, of which one is publicly available. Dice coefficient (DC) measured segmentation overlap, considering manual segmentation the ground truth. Lymphoma lesions were characterized by 31 features. Intraclass correlation coefficient (ICC) assessed features agreement between different segmentation methods. Nine hundred twenty [18F]FDG-avid lesions were identified. The SAC Bayesian method achieved the highest median intra-observer DC (0.87). Inter-observers' DC was higher for SAC Bayesian than manual segmentation (0.94 vs 0.84, p < 0.001). Semiautomatic deep learning-based median DC was promising (0.83 (Obs1), 0.79 (Obs2)). Threshold-based methods and publicly available 3D U-Net gave poorer results (0.56 ≤ DC ≤ 0.68). Maximum, mean, and peak standardized uptake values, metabolic tumor volume, and total lesion glycolysis showed excellent agreement (ICC ≥ 0.92) between manual and SAC Bayesian segmentation methods. The SAC Bayesian classifier is more reproducible and produces similar lesion features compared to manual segmentation, giving the best concordant results of all other methods. Deep learning-based segmentation can achieve overall good segmentation results but failed in few patients impacting patients' clinical evaluation.

Semi-Supervised Dual Stream Segmentation Network for Fundus Lesion Segmentation.

Accurate segmentation of retinal images can assist ophthalmologists to determine the degree of retinopathy and diagnose other systemic diseases. However, the structure of the retina is complex, and different anatomical structures often affect the segmentation of fundus lesions. In this paper, a new segmentation strategy called a dual stream segmentation network embedded into a conditional generative adversarial network is proposed to improve the accuracy of retinal lesion segmentation. First, a dual stream encoder is proposed to utilize the capabilities of two different networks and extract more feature information. Second, a multiple level fuse block is proposed to decode the richer and more effective features from the two different parallel encoders. Third, the proposed network is further trained in a semi-supervised adversarial manner to leverage from labeled images and unlabeled images with high confident pseudo labels, which are selected by the dual stream Bayesian segmentation network. An annotation discriminator is further proposed to reduce the negativity that prediction tends to become increasingly similar to the inaccurate predictions of unlabeled images. The proposed method is cross-validated in 384 clinical fundus fluorescein angiography images and 1040 optical coherence tomography images. Compared to state-of-the-art methods, the proposed method can achieve better segmentation of retinal capillary non-perfusion region and choroidal neovascularization.

12 Years of Area Variation by the Drygalski Ice Tongue as Measured With COSMO-SkyMed

One important coastal polynya around Antarctica is Terra Nova Bay (TNB) polynya. Its formation and persistence are due to the combined effect of katabatic winds, regional ice conditions, and the Drygalski Ice Tongue (DIT). The combined effect of these elements arranges a delicate balance in Terra Nova Bay. To evaluate the interacting elements, we focused our attention on the Drygalski Ice Tongue, one of the largest ice tongues in Antarctica. The DIT is a 70-kilometre long tongue which juts out like a pier from the icy land into northern McMurdo Sound of Antarctica's Ross Dependency. The analysis presented in this paper was carried out using Cosmo-SkyMed (CSK) images. The methodology developed for the analysis consists of three steps: 1) DIT shape enhancement, 2) fuzzy Bayesian segmentation which associates the analysed set of pixels with a binary label field, and 3) parametric representation of the shape of the DIT to compare the area of the ice tongue as it was 12 years ago and as it is today. The computed mean value of the DIT surface velocity was 703 m y <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> . To estimate the closeness of the derived parameter to the true value, using a statistical approach, the 95% confidence interval for the true mean value is 703±4.236 m y <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-1</sup> . Our analysis confirms a trend of advance. That is, after the significant calving events of the tongue in 2005 and 2006, the DIT has resumed its normal growth, and its length has increased by 12% in the last 12 years.

A Bayesian model based computational analysis of the relationship between bisulfite accessible single-stranded DNA in chromatin and somatic hypermutation of immunoglobulin genes.

The B cells in our body generate protective antibodies by introducing somatic hypermutations (SHM) into the variable region of immunoglobulin genes (IgVs). The mutations are generated by activation induced deaminase (AID) that converts cytosine to uracil in single stranded DNA (ssDNA) generated during transcription. Attempts have been made to correlate SHM with ssDNA using bisulfite to chemically convert cytosines that are accessible in the intact chromatin of mutating B cells. These studies have been complicated by using different definitions of “bisulfite accessible regions” (BARs). Recently, deep-sequencing has provided much larger datasets of such regions but computational methods are needed to enable this analysis. Here we leveraged the deep-sequencing approach with unique molecular identifiers and developed a novel Hidden Markov Model based Bayesian Segmentation algorithm to characterize the ssDNA regions in the IGHV4-34 gene of the human Ramos B cell line. Combining hierarchical clustering and our new Bayesian model, we identified recurrent BARs in certain subregions of both top and bottom strands of this gene. Using this new system, the average size of BARs is about 15 bp. We also identified potential G-quadruplex DNA structures in this gene and found that the BARs co-locate with G-quadruplex structures in the opposite strand. Using various correlation analyses, there is not a direct site-to-site relationship between the bisulfite accessible ssDNA and all sites of SHM but most of the highly AID mutated sites are within 15 bp of a BAR. In summary, we developed a novel platform to study single stranded DNA in chromatin at a base pair resolution that reveals potential relationships among BARs, SHM and G-quadruplexes. This platform could be applied to genome wide studies in the future.

Genotype-associated cerebellar profiles in ALS: focal cerebellar pathology and cerebro-cerebellar connectivity alterations

ObjectiveCerebellar disease burden and cerebro-cerebellar connectivity alterations are poorly characterised in amyotrophic lateral sclerosis (ALS) despite the likely contribution of cerebellar pathology to the clinical heterogeneity of the condition.MethodsA prospective...

JOINT SEGMENTATION OF MULTIPLE SCLEROSIS LESIONS AND BRAIN ANATOMY IN MRI SCANS OF ANY CONTRAST AND RESOLUTION WITH CNNs.

We present the first deep learning method to segment Multiple Sclerosis lesions and brain structures from MRI scans of any (possibly multimodal) contrast and resolution. Our method only requires segmentations to be trained (no images), as it leverages the generative model of Bayesian segmentation to generate synthetic scans with simulated lesions, which are then used to train a CNN. Our method can be retrained to segment at any resolution by adjusting the amount of synthesised partial volume. By construction, the synthetic scans are perfectly aligned with their labels, which enables training with noisy labels obtained with automatic methods. The training data are generated on the fly, and aggressive augmentation (including artefacts) is applied for improved generalisation. We demonstrate our method on two public datasets, comparing it with a state-of-the-art Bayesian approach implemented in FreeSurfer, and dataset specific CNNs trained on real data. The code is available at https://github.com/BBillot/SynthSeg.

From a Point Cloud to a Simulation Model-Bayesian Segmentation and Entropy Based Uncertainty Estimation for 3D Modelling.

The 3D modelling of indoor environments and the generation of process simulations play an important role in factory and assembly planning. In brownfield planning cases, existing data are often outdated and incomplete especially for older plants, which were mostly planned in 2D. Thus, current environment models cannot be generated directly on the basis of existing data and a holistic approach on how to build such a factory model in a highly automated fashion is mostly non-existent. Major steps in generating an environment model of a production plant include data collection, data pre-processing and object identification as well as pose estimation. In this work, we elaborate on a methodical modelling approach, which starts with the digitalization of large-scale indoor environments and ends with the generation of a static environment or simulation model. The object identification step is realized using a Bayesian neural network capable of point cloud segmentation. We elaborate on the impact of the uncertainty information estimated by a Bayesian segmentation framework on the accuracy of the generated environment model. The steps of data collection and point cloud segmentation as well as the resulting model accuracy are evaluated on a real-world data set collected at the assembly line of a large-scale automotive production plant. The Bayesian segmentation network clearly surpasses the performance of the frequentist baseline and allows us to considerably increase the accuracy of the model placement in a simulation scene.

A generalized ordinal finite mixture regression model for market segmentation

Model-based market segmentation analyses often involve an ordinal dependent variable as ordinal responses are frequently collected in marketing research. In the Bayesian segmentation literature, there are models for an interval- or ratio-scaled dependent variable but there is not any general model for an ordinal dependent variable. In this manuscript, the authors propose a new Bayesian procedure to simultaneously perform segmentation and ordinal regression with variable selection within each derived segment. The procedure is robust to outliers and it also provides an option to include concomitant variables that allows the simultaneous profiling of the derived segments. The authors demonstrate that the practice of treating ordinal responses as interval- or ratio-scales to apply existing Bayesian segmentation procedures can lead to very misleading results and conclusions. Through simulation studies, the authors show that the proposed procedure outperforms several benchmark Bayesian segmentation models in parameter recovery, segment retention, and segment membership prediction for such data. Finally, they provide a commercial business customer satisfaction empirical application to illustrate the usefulness of the proposed model.

Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes.

Single-cell transcriptomic analysis is widely used to study human tumors. However it remains challenging to distinguish normal cell types in the tumor microenvironment from malignant cells and to resolve clonal substructure within the tumor. To address these challenges, we developed an integrative Bayesian segmentation approach called CopyKAT (Copynumber Karyotyping of Aneuploid Tumors) to estimate genomic copy number profiles at an average genomic resolution of 5Mb from read depth in high-throughput scRNA-seq data. We applied CopyKAT to analyze 46,501 single cells from 21 tumors, including triple-negative breast cancer, pancreatic ductal adenocarcinomas, anaplastic thyroid cancer, invasive ductal carcinoma and glioblastoma to accurately (98%) distinguish cancer cells from normal cell types. In three breast tumors, CopyKAT resolved clonal subpopulations that differed in the expression of cancer genes such as KRAS and signatures including EMT, DNA repair, apoptosis and hypoxia. These data show that CopyKAT can aid the analysis of scRNA-seq data in a variety of solid human tumors.

An effective tumor detection approach using denoised MRI based on fuzzy bayesian segmentation approach

In the cancer treatment, the accurate segmentation of the brain tumor is a major task for which the denoised image gains major significance. This paper proposes an effective segmentation of the brain tumor for accurate detection of the brain tumors using the MRI brain image. The tumor is segmented using the proposed fuzzy Bayesian cut brain tumor segmentation approach. Initially, the noise from the MRI images is removed using the image denoising algorithm, known as Taylor-Krill Herd-based SVM algorithm. The denoised MRI output from the filter is then subjected to tumor segmentation using the proposed Fuzzy Bayesian cut brain tumor segmentation approach, which is the inclusion of the fuzzy and Gaussian naive bayes concept in the tumor cut algorithm in order to enable effective segmentation. The experimentation is performed using the BraTS database and simulated BraTS database and the comparative analysis of the proposed Fuzzy Bayesian cut brain tumor segmentation approach is performed with the other state-of-the-art method based on the metrics, such as accuracy, Jaccard similarity, Sensitivity, Specificity, and Peak signal-to-noise ratio (PSNR). The simulation results reveal that the proposed method acquired a maximum accuracy of 0.9903, by considering the Rayleigh noise using simulated BRaTS database, which is 0.20%, 0.33%, and 0.02%, better than the existing methods, such as original tumor cut, wiener filter + fuzzy Bayesian cut algorithm, and median filter + fuzzy Bayesian cut algorithm, respectively.

Abstract PO-020: Inferring copy number substructure from single-cell transcriptomics in human tumors with CopyKat

Abstract High-throughput single cell transcriptomics analysis is widely used to study human tumors, however a major challenge is distinguishing the stromal cells from the malignant cancer cells, as well as clonal substructure within tumors. To address this challenge, we developed an integrative Bayesian segmentation approach, COPYKAT to estimate genomic copy numbers at 5MB resolution from high-throughput single cell RNA-seq data. We applied COPYKAT to 39,709 single cells from 16 tumors across 4 cancer types, including premalignant and triple-negative breast cancers, pancreatic ductal adenocarcinomas, and anaplastic thyroid cancer. From these data we could accurately (98%) classify tumor cells from stromal cells. In three TNBC tumors COPYKAT resolved multiple clonal subpopulations of genotypes that differed in expression of breast cancer genes and enrichment of cancer hallmarks including EMT and hypoxia. These data show that COPYKAT can accurately resolve clonal copy number substructure in tumors and classify tumor and normal cells in a variety of human cancers. Citation Format: Ruli Gao, Shanshan Bai, Henderson Ying, Yiyun Lin, Tapsi Seth, Min Hu, Emi Sei, Alexander Davis, Fang Wang, Jennifer Rui Wang, Ken Chen, Stacey Moulder, Stephen Lai, Nicholas Navin. Inferring copy number substructure from single-cell transcriptomics in human tumors with CopyKat [abstract]. In: Proceedings of the AACR Virtual Special Conference on Tumor Heterogeneity: From Single Cells to Clinical Impact; 2020 Sep 17-18. Philadelphia (PA): AACR; Cancer Res 2020;80(21 Suppl):Abstract nr PO-020.

Crop Row Detection through UAV Surveys to Optimize On-Farm Irrigation Management

Climate change and competition among water users are increasingly leading to a reduction of water availability for irrigation; at the same time, traditionally non-irrigated crops require irrigation to achieve high quality standards. In the context of precision agriculture, particular attention is given to the optimization of on-farm irrigation management, based on the knowledge of within-field variability of crop and soil properties, to increase crop yield quality and ensure an efficient water use. Unmanned Aerial Vehicle (UAV) imagery is used in precision agriculture to monitor crop variability, but in the case of row-crops, image post-processing is required to separate crop rows from soil background and weeds. This study focuses on the crop row detection and extraction from images acquired through a UAV during the cropping season of 2018. Thresholding algorithms, classification algorithms, and Bayesian segmentation are tested and compared on three different crop types, namely grapevine, pear, and tomato, for analyzing the suitability of these methods with respect to the characteristics of each crop. The obtained results are promising, with overall accuracy greater than 90% and producer’s accuracy over 85% for the class “crop canopy”. The methods’ performances vary according to the crop types, input data, and parameters used. Some important outcomes can be pointed out from our study: NIR information does not give any particular added value, and RGB sensors should be preferred to identify crop rows; the presence of shadows in the inter-row distances may affect crop detection on vineyards. Finally, the best methodologies to be adopted for practical applications are discussed.