Machine Learning-based Segmentation Research Articles

Identification of isoAsp7-Aβ as a major Aβ variant in Alzheimer's disease, dementia with Lewy bodies and vascular dementia.

The formation of amyloid-β (Aβ) aggregates in brain is a neuropathological hallmark of Alzheimer's disease (AD). However, there is mounting evidence that Aβ also plays a pathogenic role in other types of dementia and that specific post-translational Aβ modifications contribute to its pathogenic profile. The objective of this study was to test the hypothesis that distinct types of dementia are characterized by specific patterns of post-translationally modified Aβ variants. We conducted a comparative analysis and quantified Aβ as well as Aβ with pyroglutamate (pGlu3-Aβ and pGlu11-Aβ), N-truncation (Aβ(4-X)), isoaspartate racemization (isoAsp7-Aβ and isoAsp27-Aβ), phosphorylation (pSer8-Aβ and pSer26-Aβ) or nitration (3NTyr10-Aβ) modification in post mortem human brain tissue from non-demented control subjects in comparison to tissue classified as pre-symptomaticAD (Pre-AD), AD, dementia with Lewy bodies and vascular dementia. Aβ modification-specific immunohistochemical labelings of brain sections from the posterior superior temporal gyrus were examined by machine learning-based segmentation protocols and immunoassay analyses in brain tissue after sequential Aβ extraction were carried out. Our findings revealed that AD cases displayed the highest concentrations of all Aβ variants followed by dementia with Lewy bodies, Pre-AD, vascular dementia and non-demented controls. With both analytical methods, we identified the isoAsp7-Aβ variant as a highly abundant Aβ form in all clinical conditions, followed by Aβ(4-X), pGlu3-Aβ, pGlu11-Aβ and pSer8-Aβ. These Aβ variants were detected in distinct plaque types of compact, coarse-grained, cored and diffuse morphologies and, with varying frequencies, in cerebral blood vessels. The 3NTyr10-Aβ, pSer26-Aβ and isoAsp27-Aβ variants were not found to be present in Aβ plaques but were detected intraneuronally. There was a strong positive correlation between isoAsp7-Aβ and Thal phase and a moderate negative correlation between isoAsp7-Aβ and performance on the Mini Mental State Examination. Furthermore, the abundance of all Aβ variants was highest in APOE 3/4 carriers. In aggregation assays, the isoAsp7-Aβ, pGlu3-Aβ and pGlu11-Aβ variants showed instant fibril formation without lag phase, whereas Aβ(4-X), pSer26-Aβ and isoAsp27-Aβ did not form fibrils. We conclude that targeting Aβ post-translational modifications, and in particular the highly abundant isoAsp7-Aβ variant, might be considered for diagnostic and therapeutic approaches in different types of dementia. Hence, our findings might have implications for current antibody-based therapies of AD.

Open-Source Software for Satellite-Based Crop Health Monitoring

Abstract Purpose Timely information on agricultural crops is essential to support informed crop management decisions. Satellite remote sensing technology offers a new paradigm for monitoring agricultural crops at large scale. Daily access to satellite imagery from Planet Scope allows opportunities to continuously monitor the Earth’s surface in 3-m resolution with various spectrums. However, satellites on agricultural applications are still in less information to agriculture producers due to the limited access to analytic software. The paper aims to fill the gap between the satellite resources and the agricultural users, specifically to develop satellite image analytic software publicly available for agricultural crop health monitoring. Methods Open-source software, iCalendar, was developed for high-throughput image analysis and visualization of a vegetation index (VI) and a time-series crop growth. Application programing interface (API) was used to automate the image queuing, activation, and download based on user-defined time and area of interest. Results The API was refined to filter unusable data such as multiples of clipped images and duplicates of tile images. iCalendar resulted in the seamless data processing and automated user-friendly visualization of a seasonal VI profile and a corresponding field map calendar, that are key sources of crop production cycle. Machine learning-based segmentation algorithm and radiometric calibration showed the detection and correction of abnormalities in the image. Conclusion The satellite image-based plant monitoring system through open-source software iCalendar provided a significant improvement to spatiotemporal field observation and allows growers and researchers in-season crop health assessment, thereby promoting new management strategies for crop productivity and protection.

Varroa Mite Counting Based on Hyperspectral Imaging.

Varroa mite infestation poses a severe threat to honeybee colonies globally. This study investigates the feasibility of utilizing the HS-Cam and machine learning techniques for Varroa mite counting. The methodology involves image acquisition, dimensionality reduction through Principal Component Analysis (PCA), and machine learning-based segmentation and classification algorithms. Specifically, a k-Nearest Neighbors (kNNs) model distinguishes Varroa mites from other objects in the images, while a Support Vector Machine (SVM) classifier enhances shape detection. The final phase integrates a dedicated counting algorithm, leveraging outputs from the SVM classifier to quantify Varroa mite populations in hyperspectral images. The preliminary results demonstrate segmentation accuracy exceeding 99% and an average precision of 0.9983 and recall of 0.9947 across all the classes. The results obtained from our machine learning-based approach for Varroa mite counting were compared against ground-truth labels obtained through manual counting, demonstrating a high degree of agreement between the automated counting and manual ground truth. Despite working with a limited dataset, the HS-Cam showcases its potential for Varroa counting, delivering superior performance compared to traditional RGB images. Future research directions include validating the proposed hyperspectral imaging methodology with a more extensive and diverse dataset. Additionally, the effectiveness of using a near-infrared (NIR) excitation source for Varroa detection will be explored, along with assessing smartphone integration feasibility.

A scalable and modular computational pipeline for axonal connectomics: automated tracing and assembly of axons across serial sections.

Progress in histological methods and in microscope technology has enabled dense staining and imaging of axons over large brain volumes, but tracing axons over such volumes requires new computational tools for 3D reconstruction of data acquired from serial sections. We have developed a computational pipeline for automated tracing and volume assembly of densely stained axons imaged over serial sections, which leverages machine learning-based segmentation to enable stitching and alignment with the axon traces themselves. We validated this segmentation-driven approach to volume assembly and alignment of individual axons over centimeter-scale serial sections and show the application of the output traces for analysis of local orientation and for proofreading over aligned volumes. The pipeline is scalable, and combined with recent advances in experimental approaches, should enable new studies of mesoscale connectivity and function over the whole human brain.

Volume microscopic analysis of membrane contact sites in mouse kidney renal proximal tubule epithelial cells

Subcellular organelles communicate with one another and with the plasma membrane (PM). A wide variety of membrane contact sites (MCS) tether subcellular structures to one another and mediate inter-organelle communication and exchange. The nature, distribution and function of membrane contact sites within the segment-specific architecture of renal epithelial cells has only just begun to be elucidated. We have used advanced imaging techniques, including Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) and machine learning-based segmentation, to create a comprehensive, high resolution, 3D and quantifiable volume microscopic atlas of the ultrastructure of proximal tubule epithelial cells of the mouse kidney. We applied this data set to identify and quantify MCS between the endoplasmic reticulum and mitochondria and the PM in these cells. FIB-SEM was performed on a 50 μm by 50 μm by 44 μm block of tissue prepared from the renal cortex of a perfusion fixed 12-week-old male mouse. OsO4-stained tissue was embedded in Durcupan and a 3D X-ray tomogram of the block was used to select a volume of interest that included extended lengths of multiple proximal tubule (PT). The volume was imaged by a customized enhanced FIB-SEM system. The X-Y scanning resolution was set to 4-nm, and 4-nm FIB milling was used to expose surfaces for successive imaging. Generation of the complete image stack required ~11,000 cycles of scanning/FIB ablation over ~4 weeks of uninterrupted imaging. Segmentation of the resultant image set was performed with Zeiss Arivis Pro technology. The boundary box for reconstruction had the dimensions of 29.2 μm x 24.8 μm x 8 μm with a volume of 5793.280 μm3 comprising of 2000 consecutive images at an isotropic voxel size of 4 nm. Deep learning segmentation teased apart 10 cells with distinct boundaries in the PT structure. The volume of the cells in our clipped volume ranged from 8.79 μm3 to 1438.53 μm3 with a median of 45.20 μm3. In one representative proximal tubule cell the volume of mitochondria makes up 84.91 μm3 (31.62%) of the cell volume and the volume of Endoplasmic Reticulum (ER) makes up 23.92 μm3 (8.91%) of the cell volume. The 3D reconstruction and quantitative analysis revealed that there is one dominant ER object accounting for most of the ER volume in the whole cell: i.e., the largest ER object takes up 22.37 μm3 or 93.5% of the ER volume in the cell. A post deep-learning processing algorithm used 3D object modifications to identify 11,109 3D MCS between mitochondria and ER in the 10 cells. The Mito-ER MCS ranged from 512 nm3 to 354,304 nm3 with a median volume of 1792 nm3. Visually we can see Mito-ER and Mito-PM MCS extending across a region of up to ~45 nm between the two organelles. Taken together, these data show that the basal lateral surfaces of PT cell PMs are generously invested with fenestrated smooth ER, as are mitochondria and peroxisomes. Extensive regions of direct contact between mitochondria and PM are also observed. The biochemical identities of these contacts and their physiological functions are being actively pursued. NIDDK DK072612 NIDDK DK120534. This is the full abstract presented at the American Physiology Summit 2024 meeting and is only available in HTML format. There are no additional versions or additional content available for this abstract. Physiology was not involved in the peer review process.

Lusca: FIJI (ImageJ) based tool for automated morphological analysis of cellular and subcellular structures

The human body consists of diverse subcellular, cellular and supracellular structures. Neurons possess varying-sized projections that interact with different cellular structures leading to the development of highly complex morphologies. Aiming to enhance image analysis of complex biological forms including neurons using available FIJI (ImageJ) plugins, Lusca, an advanced open-source tool, was developed. Lusca utilizes machine learning for image segmentation with intensity and size thresholds. It performs particle analysis to ascertain parameters such as area/volume, quantity, and intensity, in addition to skeletonization for determining length, branching, and width. Moreover, in conjunction with colocalization measurements, it provides an extensive set of 29 morphometric parameters for both 2D and 3D analysis. This is a significant enhancement compared to other scripts that offer only 5–15 parameters. Consequently, it ensures quicker and more precise quantification by effectively eliminating noise and discerning subtle details. With three times larger execution speed, fewer false positive and negative results, and the capacity to measure various parameters, Lusca surpasses other existing open-source solutions. Its implementation of machine learning-based segmentation facilitates versatile applications for different cell types and biological structures, including mitochondria, fibres, and vessels. Lusca’s automated and precise measurement capability makes it an ideal choice for diverse biological image analyses.

Qualitative classification of thrombus images as a way to improve quantitative analysis of thrombus formation in flow chamber assays.

Thrombus formation in vitro under flow conditions is one of the most widely used methods to study haemostasis and to evaluate the activity of potential antithrombotic compounds. Assessment of the results of these experiments is often based on a quantification of microscopic images of thrombi. In a majority of reported analysis all thrombi visualised in an image are quantified as one homogenous class. In some protocols, qualitative assessment of thrombi morphology based on a visual comparison of evaluated images with representative images of predefined classes of thrombi are performed by experienced analysts. In presented paper we show how the quantitative analysis can be improved by classification of thrombi on the basis of defined morphological features prior to quantification and we suggest that machine learning-based approach can improve this way of analysis. We tested the applicability of machine learning-based segmentation and classification of thrombi images to improve the outcome of quantification of the results of flow chamber assays. For this, we used the public domain machine learning software Ilastik for bioimage analysis developed at the European Molecular Biology Laboratory. A model was trained to distinguish two classes of thrombi based on certain morphological features which apparently correspond to the stage of thrombus development. Thrombi formed in the presence of a model antiplatelet compound-abciximab or in control conditions were quantified with the use of this model and the results were compared to quantification where all thrombi were quantified as a homogenous class. Machine learning-based analysis was capable of effective distinguishing of two classes of morphologically distinct platelet aggregates. The use of the model which segmented and quantified only the objects recognized as compacted structures provided results which better mirrored the actual effect of an antiplatelet treatment than quantification based on all structures. Classification of thrombi enabled by machine learning increases the relevance of quantitative information and allows better evaluation of the results of in vitro thrombosis assays.

Deep learning-based segmentation of subcellular organelles in high-resolution phase-contrast images.

Although quantitative analysis of biological images demands precise extraction of specific organelles or cells, it remains challenging in broad-field grayscale images, where traditional thresholding methods have been hampered due to complex image features. Nevertheless, rapidly growing artificial intelligence technology is overcoming obstacles. We previously reported the fine-tuned apodized phase-contrast microscopy system to capture high-resolution, label-free images of organelle dynamics in unstained living cells (Shimasaki, K. et al. (2024). Cell Struct. Funct., 49: 21-29). We here showed machine learning-based segmentation models for subcellular targeted objects in phase-contrast images using fluorescent markers as origins of ground truth masks. This method enables accurate segmentation of organelles in high-resolution phase-contrast images, providing a practical framework for studying cellular dynamics in unstained living cells.Key words: label-free imaging, organelle dynamics, apodized phase contrast, deep learning-based segmentation.

Machine Learning-Based Segmentation of the Thoracic Aorta with Congenital Valve Disease Using MRI.

Subjects with bicuspid aortic valves (BAV) are at risk of developing valve dysfunction and need regular clinical imaging surveillance. Management of BAV involves manual and time-consuming segmentation of the aorta for assessing left ventricular function, jet velocity, gradient, shear stress, and valve area with aortic valve stenosis. This paper aims to employ machine learning-based (ML) segmentation as a potential for improved BAV assessment and reducing manual bias. The focus is on quantifying the relationship between valve morphology and vortical structures, and analyzing how valve morphology influences the aorta's susceptibility to shear stress that may lead to valve incompetence. The ML-based segmentation that is employed is trained on whole-body Computed Tomography (CT). Magnetic Resonance Imaging (MRI) is acquired from six subjects, three with tricuspid aortic valves (TAV) and three functionally BAV, with right-left leaflet fusion. These are used for segmentation of the cardiovascular system and delineation of four-dimensional phase-contrast magnetic resonance imaging (4D-PCMRI) for quantification of vortical structures and wall shear stress. The ML-based segmentation model exhibits a high Dice score (0.86) for the heart organ, indicating a robust segmentation. However, the Dice score for the thoracic aorta is comparatively poor (0.72). It is found that wall shear stress is predominantly symmetric in TAVs. BAVs exhibit highly asymmetric wall shear stress, with the region opposite the fused coronary leaflets experiencing elevated tangential wall shear stress. This is due to the higher tangential velocity explained by helical flow, proximally of the sinutubal junction of the ascending aorta. ML-based segmentation not only reduces the runtime of assessing the hemodynamic effectiveness, but also identifies the significance of the tangential wall shear stress in addition to the axial wall shear stress that may lead to the progression of valve incompetence in BAVs, which could guide potential adjustments in surgical interventions.

Surgical navigation for guidewire placement from intraoperative fluoroscopy in orthopaedic surgery.

Objective. Surgical guidewires are commonly used in placing fixation implants to stabilize fractures. Accurate positioning of these instruments is challenged by difficulties in 3D reckoning from 2D fluoroscopy. This work aims to enhance the accuracy and reduce exposure times by providing 3D navigation for guidewire placement from as little as two fluoroscopic images.Approach. Our approach combines machine learning-based segmentation with the geometric model of the imager to determine the 3D poses of guidewires. Instrument tips are encoded as individual keypoints, and the segmentation masks are processed to estimate the trajectory. Correspondence between detections in multiple views is established using the pre-calibrated system geometry, and the corresponding features are backprojected to obtain the 3D pose. Guidewire 3D directions were computed using both an analytical and an optimization-based method. The complete approach was evaluated in cadaveric specimens with respect to potential confounding effects from the imaging geometry and radiographic scene clutter due to other instruments.Main results. The detection network identified the guidewire tips within 2.2 mm and guidewire directions within 1.1°, in 2D detector coordinates. Feature correspondence rejected false detections, particularly in images with other instruments, to achieve 83% precision and 90% recall. Estimating the 3D direction via numerical optimization showed added robustness to guidewires aligned with the gantry rotation plane. Guidewire tips and directions were localized in 3D world coordinates with a median accuracy of 1.8 mm and 2.7°, respectively.Significance. The paper reports a new method for automatic 2D detection and 3D localization of guidewires from pairs of fluoroscopic images. Localized guidewires can be virtually overlaid on the patient's pre-operative 3D scan during the intervention. Accurate pose determination for multiple guidewires from two images offers to reduce radiation dose by minimizing the need for repeated imaging and provides quantitative feedback prior to implant placement.

3D architecture and a bicellular mechanism of touch detection in mechanosensory corpuscle.

Mechanosensory corpuscles detect transient touch and vibration in the skin of vertebrates, enabling precise sensation of the physical environment. The corpuscle contains a mechanoreceptor afferent surrounded by lamellar cells (LCs), but corpuscular ultrastructure and the role of LCs in touch detection are unknown. We report the three-dimensional architecture of the avian Meissner (Grandry) corpuscle acquired using enhanced focused ion beam scanning electron microscopy and machine learning-based segmentation. The corpuscle comprises a stack of LCs interdigitated with terminal endings from two afferents. Simultaneous electrophysiological recordings from both cell types revealed that mechanosensitive LCs use calcium influx to trigger action potentials in the afferent and thus serve as physiological touch sensors in the skin. The elaborate architecture and bicellular sensory mechanism in the corpuscles, which comprises the afferents and LCs, create the capacity for nuanced encoding of the submodalities of touch.

3D microstructure characterization of Cu 25Cr solid state sintered alloy using X-ray computed tomography and machine learning assisted segmentation

Cu-Cr-based alloys with Cr content from 5 to 50 wt% are widely used as electrical contacts for vacuum interrupters for medium voltage applications because of their excellent combination of mechanical, thermal, and electrical conductivity. CuCr electrical contacts are usually processed by sintering or casting processes. For solid-state sintered CuCr materials, the physical properties vary as a function of the Cr content, phase morphology and porosity volume fraction. Some studies have investigated the effect of the microstructural characteristics of CuCr alloys with different Cr content and morphology on their properties. However, the porosity characterization and Cr spatial distribution and how they affect these alloys' physical properties are not as well documented. In this study, we report an in-depth 3D characterization of the porosity and Cr-phase of solid-state sintered Cu25Cr alloys with three final relative densities using X-ray Computed Tomography (XCT). An image analysis algorithm assisted by a machine learning-based segmentation method has been specifically developed. Results show that for Cu25Cr solid sintered alloys there are mainly two types of pores, pores located at the Cu/Cr interfaces, and pores within the Cu matrix. The interfacial porosity represents the larger volume fraction, over 75% of the total porosity for all cases, forming a large network of interconnected pores. With the increase of final density, the Cu-matrix becomes nearly fully dense while interfacial pores still represent the largest fraction decreases in size and volume. These interfacial pores networks are believed to be formed due to poor filling and packing of Cu around the percolated Cr-phase. These observations might be helpful to optimize the functional properties of CuCr sintered alloys.

Anomalous NO[formula omitted] emitting ship detection with TROPOMI satellite data and machine learning

Starting from 2021, more demanding NOx emission restrictions were introduced for ships operating in the North and Baltic Sea waters. Since all methods currently used for ship compliance monitoring are financially and time demanding, it is important to prioritize the inspection of ships that have high chances of being non-compliant. The current state-of-the-art approach for a large-scale ship NO2 estimation is a supervised machine learning-based segmentation of ship plumes on TROPOMI/S5P images. However, challenging data annotation and insufficiently complex ship emission proxy used for the validation limit the applicability of the model for ship compliance monitoring. In this study, we present a methodology towards the automated and scalable selection of potentially non-compliant ships using a combination of machine learning models on TROPOMI satellite data. It is based on a proposed regression model predicting the amount of NO2 that is expected to be produced by a ship with certain properties operating in the given atmospheric conditions. The model does not require manual labeling and is validated with TROPOMI data directly. The differences between the predicted and actual amount of produced NO2 are integrated over observations of the ship in time and are used as a measure of the inspection worthiness of a ship. To add further evidence, we compare the obtained results with the results of the previously developed segmentation-based method. Ships that are also highly deviating in accordance with the segmentation method require further attention. If no other explanations can be found by checking the TROPOMI data, the respective ships are advised to be the candidates for inspection.

Vectorize Me! A Proposed Machine Learning Approach for Segmenting the Multi-optional Tourist

Contemporary consumer behavior is characterized by its multidimensionality and complexity, which, at the same time, pushes traditional segmentation approaches to their limits. In response, this methodological study proposes a multistage machine learning-based segmentation process using semiotic-semantic community detection. This innovative method was conducted exemplarily and evaluated on a representative sample of 1,101 German travelers. The main contribution of this study lies in the novel use of word vectors, which result from assigning a semiotic meaning to travel-type images. Thus, high-dimensional data could be used during the segmentation process, overcoming several classical segmentation problems. By using semantic similarities, tourists could be grouped and represented in their multidimensionality. From a theoretical perspective, this study was inspired by postmodern tourism practices in order to better understand the (oftentimes) hybrid and multilayered behaviors of tourists. To make this innovative approach reproducible, recommendations for implementation and all necessary data have been provided.

Machine Learning-based Segmentation to the Prediction of Liver Cirrhosis

Machine Learning-based Segmentation to the Prediction of Liver Cirrhosis

Machine learning-based segmentation of the rodent hippocampal CA2 area from Nissl-stained sections.

The hippocampus is a center of learning, memory, and spatial navigation. This region is divided into the CA1, CA2, and CA3 areas, which are anatomically different from each other. Among these divisions, the CA2 area is unique in terms of functional relevance to sociality. The CA2 area is often manually detected based on the size, shape, and density of neurons in the hippocampal pyramidal cell layer, but this manual segmentation relying on cytoarchitecture is impractical to apply to a large number of samples and dependent on experimenters' proficiency. Moreover, the CA2 area has been defined based on expression pattern of molecular marker proteins, but it generally takes days to complete immunostaining for such proteins. Thus, we asked whether the CA2 area can be systematically segmented based on cytoarchitecture alone. Since the expression pattern of regulator of G-protein signaling 14 (RGS14) signifies the CA2 area, we visualized the CA2 area in the mouse hippocampus by RGS14-immunostaining and Nissl-counterstaining and manually delineated the CA2 area. We then established "CAseg," a machine learning-based automated algorithm to segment the CA2 area with the F1-score of approximately 0.8 solely from Nissl-counterstained images that visualized cytoarchitecture. CAseg was extended to the segmentation of the prairie vole CA2 area, which raises the possibility that the use of this algorithm can be expanded to other species. Thus, CAseg will be beneficial for investigating unique properties of the hippocampal CA2 area.

Drug screening at single-organoid resolution via bioprinting and interferometry

High throughput drug screening is an established approach to investigate tumor biology and identify therapeutic leads. Traditional platforms use two-dimensional cultures which do not accurately reflect the biology of human tumors. More clinically relevant model systems such as three-dimensional tumor organoids can be difficult to scale and screen. Manually seeded organoids coupled to destructive endpoint assays allow for the characterization of treatment response, but do not capture transitory changes and intra-sample heterogeneity underlying clinically observed resistance to therapy. We present a pipeline to generate bioprinted tumor organoids linked to label-free, time-resolved imaging via high-speed live cell interferometry (HSLCI) and machine learning-based quantitation of individual organoids. Bioprinting cells gives rise to 3D structures with unaltered tumor histology and gene expression profiles. HSLCI imaging in tandem with machine learning-based segmentation and classification tools enables accurate, label-free parallel mass measurements for thousands of organoids. We demonstrate that this strategy identifies organoids transiently or persistently sensitive or resistant to specific therapies, information that could be used to guide rapid therapy selection.

RETRACTED ARTICLE: Machine learning-based segmentation of images to diagnose the orthopedic diseases and to guide the orthopedic surgeries

RETRACTED ARTICLE: Machine learning-based segmentation of images to diagnose the orthopedic diseases and to guide the orthopedic surgeries

Semi-supervised nuclei segmentation based on multi-edge features fusion attention network.

The morphology of the nuclei represents most of the clinical pathological information, and nuclei segmentation is a vital step in current automated histopathological image analysis. Supervised machine learning-based segmentation models have already achieved outstanding performance with sufficiently precise human annotations. Nevertheless, outlining such labels on numerous nuclei is extremely professional needing and time consuming. Automatic nuclei segmentation with minimal manual interventions is highly needed to promote the effectiveness of clinical pathological researches. Semi-supervised learning greatly reduces the dependence on labeled samples while ensuring sufficient accuracy. In this paper, we propose a Multi-Edge Feature Fusion Attention Network (MEFFA-Net) with three feature inputs including image, pseudo-mask and edge, which enhances its learning ability by considering multiple features. Only a few labeled nuclei boundaries are used to train annotations on the remaining mostly unlabeled data. The MEFFA-Net creates more precise boundary masks for nucleus segmentation based on pseudo-masks, which greatly reduces the dependence on manual labeling. The MEFFA-Block focuses on the nuclei outline and selects features conducive to segment, making full use of the multiple features in segmentation. Experimental results on public multi-organ databases including MoNuSeg, CPM-17 and CoNSeP show that the proposed model has the mean IoU segmentation evaluations of 0.706, 0.751, and 0.722, respectively. The model also achieves better results than some cutting-edge methods while the labeling work is reduced to 1/8 of common supervised strategies. Our method provides a more efficient and accurate basis for nuclei segmentations and further quantifications in pathological researches.

Nailfold Capillaroscopy: A Promising, Noninvasive Approach to Predict Retinopathy of Prematurity

To test the hypothesis that nailfold capillaroscopy can noninvasively detect dysregulated retinal angiogenesis and predict retinopathy of prematurity (ROP) in infants born premature before its development. In a cohort of 32 infants born <33weeks of gestation, 1386 nailfold capillary network images of the 3 middle fingers of each hand were taken during the first month of life. From these, 25 infants had paired data taken 2 weeks apart during the first month of life. Images were analyzed for metrics of peripheral microvascular density using a machine learning-based segmentation approach and a previously validated microvascular quantification platform (REAVER vascular analysis). Results were correlated with subsequent development of ROP based on a published consensus ROP severity scale. In total, 18 of 32 (56%) (entire cohort) and 13 of 25 (52%) (2-time point subgroup) developed ROP. Peripheral vascular density decreased significantly during the first month of life. In the paired time point analysis, vessel length density, a key metric of peripheral vascular density, was significantly greater at both time points among infants who later developed ROP (15 563 and 11 996μm/mm2, respectively) compared with infants who did not (12 252 and 8845μm/mm2, respectively) (P<.001, both time points). A vessel length density cutoff of >15 100 at T1 or at T2 correctly detected 3 of 3 infants requiring ROP therapy. In a mixed-effects linear regression model, peripheral vascular density metrics were significantly correlated with ROP severity. Nailfold microvascular density assessed during the first month of life is a promising, noninvasive biomarker to identify premature infants at highest risk for ROP before detection on eye exam.