Automatic Annotation Algorithm Research Articles

SIVED: A SAR Image Dataset for Vehicle Detection Based on Rotatable Bounding Box

The research and development of deep learning methods are heavily reliant on large datasets, and there is currently a lack of scene-rich datasets for synthetic aperture radar (SAR) image vehicle detection. To address this issue and promote the development of SAR vehicle detection algorithms, we constructed the SAR Image dataset for VEhicle Detection (SIVED) using Ka, Ku, and X bands of data. Rotatable bounding box annotations were employed to improve positioning accuracy, and an algorithm for automatic annotation was proposed to improve efficiency. The dataset exhibits three crucial properties: richness, stability, and challenge. It comprises 1044 chips and 12,013 vehicle instances, most of which are situated in complex backgrounds. To construct a baseline, eight detection algorithms are evaluated on SIVED. The experimental results show that all detectors achieved high mean average precision (mAP) on the test set, highlighting the dataset’s stability. However, there is still room for improvement in the accuracy with respect to the complexity of the background. In summary, SIVED fills the gap in SAR image vehicle detection datasets and demonstrates good adaptability for the development of deep learning algorithms.

An automatic facial landmarking for children with rare diseases.

Two to three thousand syndromes modify facial features: their screening requires the eye of an expert in dysmorphology. A widely used tool in shape characterization is geometric morphometrics based on landmarks, which are precise and reproducible anatomical points. Landmark positioning is user dependent and time consuming. Many automatic landmarking tools are currently available but do not work for children, because they have mainly been trained using photographic databases of healthy adults. Here, we developed a method for building an automatic landmarking pipeline for frontal and lateral facial photographs as well as photographs of external ears. We evaluated the algorithm on patients diagnosed with Treacher Collins (TC) syndrome as it is the most frequent mandibulofacial dysostosis in humans and is clinically recognizable although highly variable in severity. We extracted photographs from the photographic database of the maxillofacial surgery and plastic surgery department of Hôpital Necker-Enfants Malades in Paris, France with the diagnosis of TC syndrome. The control group was built from children admitted for craniofacial trauma or skin lesions. After testing two methods of object detection by bounding boxes, a Haar Cascade-based tool and a Faster Region-based Convolutional Neural Network (Faster R-CNN)-based tool, we evaluated three different automatic annotation algorithms: the patch-based active appearance model (AAM), the holistic AAM, and the constrained local model (CLM). The final error corresponding to the distance between the points placed by automatic annotation and those placed by manual annotation was reported. We included, respectively, 1664, 2044, and 1375 manually annotated frontal, profile, and ear photographs. Object recognition was optimized with the Faster R-CNN-based detector. The best annotation model was the patch-based AAM (p < 0.001 for frontal faces, p=0.082 for profile faces and p < 0.001 for ears). This automatic annotation model resulted in the same classification performance as manually annotated data. Pretraining on public photographs did not improve the performance of the model. We defined a pipeline to create automatic annotation models adapted to faces with congenital anomalies, an essential prerequisite for research in dysmorphology.

Beat-Level Interpretation of Intra-Patient Paradigm Based on Object Detection.

Electrocardiogram (ECG), as a product that can most directly reflect the electrical activity of the heart, has become the most common clinical technique used for the analysis of cardiac abnormalities. However, it is a heavy and tedious burden for doctors to analyze a large amount of ECG data from the long-term monitoring system. The realization of automatic ECG analysis is of great significance. This work proposes a beat-level interpretation method based on the automatic annotation algorithm and object detector, which abandons the previous mode of separate R peak detection and heartbeat classification. The ground truth of the QRS complex is automatically annotated and also regarded as the object the model can learn like category information. The object detector unifies the localization and classification tasks, achieving an end-to-end optimization as well as decoupling the high dependence on the R peak. Compared with most advanced methods, this work shows superior performance. For the interpretation of 12 heartbeat types in the MIT-BIH dataset, the average accuracy is 99.60%, the average sensitivity is 97.56%, and the average specificity is 99.78%. This method can be used as a clinical auxiliary tool to help doctors diagnose arrhythmia after receiving large-scale database training.

Weighted IForest and siamese GRU on small sample anomaly detection in healthcare

Background and objectiveAt present, many achievements have been made in anomaly detection of big data using deep neural network, However, in many practical application scenarios, there are still some problems, such as shortage of data, too large workload of manual data annotating and so on.MethodsThis paper proposes weighted iForest and Siamese GRU (WIF-SGRU) algorithm on small sample anomaly detection. In the data annotation stage, we propose a weighted IForest algorithm for automatic annotation of unlabeled data. In the training phase of anomaly detection model, the Siamese GRU is proposed to train the target data to obtain the anomaly model and detect the real-time anomaly of small sample data.ResultsThe proposed algorithm is verified on six public datasets (Arrhythmia, Shuttle, Staellite, Sttimage-2, Lymphography, and WBC). The experimental results show that compared with the traditional data annotation and anomaly detection algorithm, the algorithm of weighted IForest and Siamese GRU improves the accuracy and real-time performance.ConclusionsThis paper proposes a weighted IForest and Siamese GRU algorithm architecture, which provides a more accurate and efficient method for outlier detection of data. Firstly, the framework uses the improved IForest algorithm to label the label-free data, Then the Siamese GRU is optimized by the improved FDAloss function,the optimized network is used to learn the distance between data for real-time and efficient anomaly detection. Experiments show that the framework has good potential.

Automatic Annotation Algorithm of Medical Radiological Images using Convolutional Neural Network

In order to address the problems of time-consuming, low accuracy and poor convergence effect of traditional image automatic annotation algorithm, an Automatic annotation algorithm of medical radiological images based on convolutional neural network (CNN) is proposed. First of all, the image gradient information model was constructed, the edge contour feature of medical radiation image was initialized, the automatic segmentation model of medical radiation image was established by block template matching method, and the automatic segmentation processing of medical radiation image was completed. Secondly, by fusing the contour and gray information of image segmentation, the multi-resolution feature is extracted by using the three-dimensional distributed pixel sequence of image. The fusion feature decomposition of the image was obtained based on CNN, and the automatic annotation of medical radiation image was completed. The results show that the image segmentation effect of the proposed algorithm is good, the number of feature points is accurate, and the accuracy of multi-resolution feature extraction is as high as 98.7%. The convergence of image annotation is good, short time-consumption, and the F1 measurement value of the algorithm is high, and the overall performance is good.

Outcome of catheter ablation of arrhythmias in adult congenital heart disease with high density multi-electrode mapping with automatic annotation algorithm

Abstract Funding Acknowledgements Type of funding sources: None. Introduction Arrhythmias are frequently encountered in adult congenital heart disease (ACHD) and respond poorly to pharmacological therapies. Catheter ablation is challenging due to anatomical variation and complexity of the arrhythmia substrate. High density multi electrode mapping (MEM) with automatic annotation of activation time may aid mapping of arrhythmia, decrease procedure time and improve the accuracy of targeting of ablation therapy. Purpose To compare the acute and long term outcomes and procedural characteristics of catheter ablation in ACHD patients with and without automatic annotation of activation with MEM. Methods Retrospective analysis of the acute and long term outcomes of ACHD patients in a single centre undergoing ablation procedures from 1 Jan 2014 to 18 August 2017 was undertaken. 2 groups were identified. Group 1 included patients who had arrhythmia mapping performed with the CARTO 3D electroanatomic mapping system without the use of automatic signal annotation. Most patients in this group had sequential mapping performed with the ablation catheter (78%), the rest had multi-electrode mapping with the PentaRay 20 pole catheter. Group 2 included patients who had arrhythmia mapping performed with the CARTO 3D electroanatomic mapping system using the automated CONFIDENSE mapping algorithm. Results Group 1: n = 27, mean age 44.6 +/-3 years. Male 46.6%. Group 2: n = 38, mean age 44.0 +/- 1.9 years. Male 56.7%. All patients had CHD of at least moderate complexity. 25% of patients in group 1 and 45% in group 2 were repeat ablations. 45 arrhythmias were induced in group 1 of which 29 were targeted and 74 arrhythmias were induced in group 2 of which 46 were targeted. Acute success rates (after attempts at reinduction) were 96.3% in group 1 and 94.7% in group 2. Recurrences of arrhythmia occurred significantly less in patients in group 2 compared to group 1 (44.7% and 70.4% respectively, p &amp;lt; 0.05) after a follow up duration of 17.3+/-0.43 months in group 2 and 45.3 +/-1.19 months in group 1. Fluoroscopy time, procedure time and ablation time were not significantly different between groups. Conclusions The use of multi-electrode mapping with an automatic annotation algorithm was associated with a significantly lower risk of recurrence during the follow up period of this study.

Quantification of immobilization-induced changes in human calf muscle using speed-of-sound ultrasound: An observational pilot study.

Short-term immobilization leads to fatty muscular degeneration, which is associated with various negative health effects. Based on literature showing very high correlations between MRI Dixon fat fraction and Speed-of-Sound (SoS), we hypothesized that we can detect short-term-immobilization-induced differences in SoS.Both calves of 10 patients with a calf cast on one side for a mean duration of 41 ± 26 days were examined in relaxed position using a standard ultrasound machine. Calf perimeters were measured for both sides. A flat Plexiglas-reflector, placed vertically on the opposite side of the probe with the calf in-between, was used as a timing reference for SoS. SoS was both manually annotated by two readers and assessed by an automatic annotation algorithm. The thickness values of the subcutaneous fat and muscle layers were manually read from the B-mode images. Differences between the cast and non-cast calves were calculated with a paired t test. Correlation analysis of SoS and calf perimeter was performed using Pearson's correlation coefficient.Paired t test showed significant differences between the cast and non-cast side for both SoS (P < .01) and leg perimeter (P < .001). SoS was reduced with the number of days after cast installment (r = −0.553, P = .097). No significant differences were found for muscle layer thickness, subcutaneous fat layer thickness, mean fat echo intensity, or mean muscle echo intensity.Short-term-immobilization led to a significant reduction in SoS in the cast calf compared to the healthy calf, indicating a potential role of SoS as a biomarker in detecting immobilization-induced fatty muscular degeneration not visible on B-mode ultrasound.

Quantitation of Molecular Pathway Activation Using RNA Sequencing Data.

Intracellular molecular pathways (IMPs) control all major events in the living cell. IMPs are considered hotspots in biomedical sciences and thousands of IMPs have been discovered for humans and model organisms. Knowledge of IMPs activation is essential for understanding biological functions and differences between the biological objects at the molecular level. Here we describe the Oncobox system for accurate quantitative scoring activities of up to several thousand molecular pathways based on high throughput molecular data. Although initially designed for gene expression and mainly RNA sequencing data, Oncobox is now also applicable for quantitative proteomics, microRNA and transcription factor binding sites mapping data. The Oncobox system includes modules of gene expression data harmonization, aggregation and comparison and a recursive algorithm for automatic annotation of molecular pathways. The universal rationale of Oncobox enables scoring of signaling, metabolic, cytoskeleton, immunity, DNA repair, and other pathways in a multitude of biological objects. The Oncobox system can be helpful to all those working in the fields of genetics, biochemistry, interactomics, and big data analytics in molecular biomedicine.

Repertoire-wide gene structure analyses: a case study comparing automatically predicted and manually annotated gene models

BackgroundThe location and modular structure of eukaryotic protein-coding genes in genomic sequences can be automatically predicted by gene annotation algorithms. These predictions are often used for comparative studies on gene structure, gene repertoires, and genome evolution. However, automatic annotation algorithms do not yet correctly identify all genes within a genome, and manual annotation is often necessary to obtain accurate gene models and gene sets. As manual annotation is time-consuming, only a fraction of the gene models in a genome is typically manually annotated, and this fraction often differs between species. To assess the impact of manual annotation efforts on genome-wide analyses of gene structural properties, we compared the structural properties of protein-coding genes in seven diverse insect species sequenced by the i5k initiative.ResultsOur results show that the subset of genes chosen for manual annotation by a research community (3.5–7% of gene models) may have structural properties (e.g., lengths and exon counts) that are not necessarily representative for a species’ gene set as a whole. Nonetheless, the structural properties of automatically generated gene models are only altered marginally (if at all) through manual annotation. Major correlative trends, for example a negative correlation between genome size and exonic proportion, can be inferred from either the automatically predicted or manually annotated gene models alike. Vice versa, some previously reported trends did not appear in either the automatic or manually annotated gene sets, pointing towards insect-specific gene structural peculiarities.ConclusionsIn our analysis of gene structural properties, automatically predicted gene models proved to be sufficiently reliable to recover the same gene-repertoire-wide correlative trends that we found when focusing on manually annotated gene models only. We acknowledge that analyses on the individual gene level clearly benefit from manual curation. However, as genome sequencing and annotation projects often differ in the extent of their manual annotation and curation efforts, our results indicate that comparative studies analyzing gene structural properties in these genomes can nonetheless be justifiable and informative.

Manual semantic annotations: User evaluation of interface and interaction designs

Semantic annotation is the process by which existing texts receive a mark-up that allows automatic identification of named entities (e.g., to distinguish between a turkey bird and the country Turkey). Manual annotation is useful both as stand-alone process in a domain-specific setting or as post-processing for automatic annotation algorithms. Development of most annotation tools strongly focuses on providing novel functionality — end-user evaluations of interface and user interaction are rare.This article reports on the results of a series of user studies executed to explore how non-expert users understand the process of enriching texts with semantic annotations. We find that these users can easily create simple semantic annotations (e.g., assigning concepts to text passages) but have difficulties understanding complex semantic annotations (e.g., assigning semantic identifiers to text passages). We present the results of three user studies on manual semantic annotation and discusses the lessons learned about both the semantic enrichment process and our methodology of exposing non-experts to semantic enrichment.

RNA sequencing analysis for profiling activation of cancer-associated molecular pathways.

e13032 Background: Intracellular molecular pathways (IMPs) control all major events in the living cell. They are considered hotspots in contemporary oncology because knowledge of IMPs activation is essential for understanding mechanisms of molecular pathogenesis in oncology. Profiling IMPs requires RNA-seq data for tumors and for a collection of reference normal tissues. However, there is a shortage now in such profiles for normal tissues from healthy human donors, uniformly profiled in a single series of experiments. Access to the largest dataset of normal profiles GTEx is only partly available through the dbGaP. In TCGA database, norms are adjacent to surgically removed tumors and may be affected by tumor-linked growth factors, inflammation and altered vascularization. ENCODE datasets were for the autopsies of normal tissues, but they can’t form statistically significant reference groups. Methods: Tissue samples representing 20 organs were taken from post-mortal human healthy donors killed in road accidents no later than 36 hours after death, blood samples were taken from healthy volunteers. Gene expression was profiled in RNA-seq experiments using the same reagents, equipment and protocols. Bioinformatic algorithms for IMP analysis were developed and validated using experimental and public gene expression datasets. Results: From original sequencing data we constructed the biggest fully open reference expression database of normal human tissues including 465 profiles termed Oncobox Atlas of Normal Tissue Expression (ANTE, original data: GSE120795). We next developed a method termed Oncobox for interrogating activation of IMPs in human cancers. It includes modules of expression data harmonization and comparison and an algorithm for automatic annotation of molecular pathways. The Oncobox system enables accurate scoring of thousands molecular pathways using RNA-seq data. Oncobox pathway analysis is also applicable for quantitative proteomics and microRNA data in oncology. Conclusions: The Oncobox system can be used for a plethora of applications in cancer research including finding differentially regulated genes and IMPs, and for discovery of new pathway-related diagnostic and prognostic biomarkers.

Deep Neural Network Method of Recognizing the Critical Situations for Transport Systems by Video Images

The deep neural network method of recognizing critical situations for transport systems according to video frames from the intelligent vehicles cameras is offered, that is effective in terms of accuracy and high-speed performance. Unlike the known solutions for the objects and normal or critical situations detection and recognition, it uses the classification with the subsequent reinforcement on the basis of several video stream frames and with the automatic annotation algorithm. The adapted architectures of neural networks are offered: the dual network to identify drivers and passengers according to the face image, the network with independent recurrent layers to classify situations according to the video fragment. The scheme of the intellectual distributed city system of transport safety using the cameras and on-board computers united in a single network is offered. Software modules in Python are developed and natural experiments are made. The possibility of the offered algorithms and programs in UGV or in the driver assistant systems implementation is shown with the illustrating examples in real-time.

An Automatic Annotation Algorithm for Deep Learning Image Datasets Based on HOG Features

Labeled data is crucial for deep learning model training, but it is mainly based on manual annotation. This paper proposes an automatic annotation algorithm for deep learning image datasets based on HOG features. The algorithm obtains the position information of the target based on the Gaussian mixture model, and automatically obtains the target category based on the Mean Shift algorithm of HOG feature. The labeled sample obtained in the above steps is used to train the Deformable Part Model to annotate more unlabeled image data. Experimental results show that this algorithm can quickly complete the automatic annotation of image datasets, reduce the cost of acquiring labeled datasets and greatly improve the effectiveness of acquiring labeled datasets in deep learning.

AUTOMATIC MUSCLE PERIMYSIUM ANNOTATION USING DEEP CONVOLUTIONAL NEURAL NETWORK.

Diseased skeletal muscle expresses mononuclear cell infiltration in the regions of perimysium. Accurate annotation or segmentation of perimysium can help biologists and clinicians to determine individualized patient treatment and allow for reasonable prognostication. However, manual perimysium annotation is time consuming and prone to inter-observer variations. Meanwhile, the presence of ambiguous patterns in muscle images significantly challenge many traditional automatic annotation algorithms. In this paper, we propose an automatic perimysium annotation algorithm based on deep convolutional neural network (CNN). We formulate the automatic annotation of perimysium in muscle images as a pixel-wise classification problem, and the CNN is trained to label each image pixel with raw RGB values of the patch centered at the pixel. The algorithm is applied to 82 diseased skeletal muscle images. We have achieved an average precision of 94% on the test dataset.

Graph-Based Label Propagation in Digital Media

The expansion of the Internet over the last decade and the proliferation of online social communities, such as Facebook, Google+, and Twitter, as well as multimedia sharing sites, such as YouTube, Flickr, and Picasa, has led to a vast increase of available information to the user. In the case of multimedia data, such as images and videos, fast querying and processing of the available information requires the annotation of the multimedia data with semantic descriptors, that is, labels. However, only a small proportion of the available data are labeled. The rest should undergo an annotation-labeling process. The necessity for the creation of automatic annotation algorithms gave birth to label propagation and semi-supervised learning. In this study, basic concepts in graph-based label propagation methods are discussed. Methods for proper graph construction based on the structure of the available data and label inference methods for spreading label information from a few labeled data to a larger set of unlabeled data are reviewed. Applications of label propagation algorithms in digital media, as well as evaluation metrics for measuring their performance, are presented.