True Positive Detection Research Articles

Multispecies bioacoustics classification using transfer learning of deep convolutional neural networks with pseudo-labeling

In this study, we evaluated deep convolutional neural networks for classifying the calls of 24 birds and amphibian species detected in ambient field recordings from the tropical mountains of Puerto Rico. Training data were collected using a template-based detection algorithm and manually validated with a graphical interface. To reduce the labor intensive and time-consuming process of manual validation, as well as to increase the accuracy of species classification with acoustic recordings, we propose a novel approach that combines transfer learning of a pre-trained deep convolutional neural network (CNN) model, a semi-supervised pseudo-labeling method, and a custom training loss function. While generating sufficient training data is a major challenge for many deep learning applications, our proposed methodology enables the network to be trained in a supervised fashion with labeled and unlabeled data simultaneously, which effectively increases the size of training set and thus boosts the model performance. As a result, the model achieves 97.7% sensitivity, 96.4% specificity and 96.6% accuracy in classifying a test set of manually validated true and false positive template-based detections. This multi-label multi-species classification methodology and its framework can be easily expanded to other acoustic classification problems.

Convolutional neural networks for improving image quality with noisy PET data

GoalPET is a relatively noisy process compared to other imaging modalities, and sparsity of acquisition data leads to noise in the images. Recent work has focused on machine learning techniques to improve PET images, and this study investigates a deep learning approach to improve the quality of reconstructed image volumes through denoising by a 3D convolution neural network. Potential improvements were evaluated within a clinical context by physician performance in a reading task.MethodsA wide range of controlled noise levels was emulated from a set of chest PET data in patients with lung cancer, and a convolutional neural network was trained to denoise the reconstructed images using the full-count reconstructions as the ground truth. The benefits, over conventional Gaussian smoothing, were quantified across all noise levels by observer performance in an image ranking and lesion detection task.ResultsThe CNN-denoised images were generally ranked by the physicians equal to or better than the Gaussian-smoothed images for all count levels, with the largest effects observed in the lowest-count image sets. For the CNN-denoised images, overall lesion contrast recovery was 60% and 90% at the 1 and 20 million count levels, respectively. Notwithstanding the reduced lesion contrast recovery in noisy data, the CNN-denoised images also yielded better lesion detectability in low count levels. For example, at 1 million true counts, the average true positive detection rate was around 40% for the CNN-denoised images and 30% for the smoothed images.ConclusionSignificant improvements were found for CNN-denoising for very noisy images, and to some degree for all noise levels. The technique presented here offered however limited benefit for detection performance for images at the count levels routinely encountered in the clinic.

Real-time identification of aircraft sound events

Metropolitan airports constitute an environmental nuisance, mainly due to noise pollution originating from aircraft landings and takeoffs, affecting the wellbeing of the airports’ neighboring populations. Noise measurement is considered the fundamental means to evaluate, enforce, validate, and control noise abatement. Noise measurements performed by sound monitors located close to urban airports are often disrupted by urban background noise that interferes with aircraft sounds. Detecting aircraft noise, classifying, identifying, and separating it from the residual background noise is a challenge for unattended aircraft noise monitors. This paper suggests a simple and inexpensive methodology, based on ADS-B (Automatic Dependent Surveillance-Broadcast), which can facilitate isolating aircraft noise from background noise. Experiments showed that using ADS-B driven noise monitors is at least as accurate as the commonly used radar–driven noise monitors, in terms of true positive, false positive, or false negative detection during the examined periods.

A Trust-Management-Based Intrusion Detection System for Routing Protocol Attacks in Internet of Things

Abstract— The Internet of things is a pool of on-demand and configurable resources and services that are delivered across the usage of the internet. Providing privacy and security to protect their resources is considered a very challenging issue since the distributed architecture of the cloud makes it vulnerable to the intruders. To mitigate this issue, intrusion detection system plays an important role in detecting the attacks in the network. Intrusion detection system is a software or hardware component that implements monitoring and analysis processes of the system events or network activities. Once detecting any intrusion, an alert is raised to the administrator in order to take appropriate actions against such these intrusive events. In this paper an intrusion detection system is proposed for routing protocol for lossy and low power network attacks. The objective of the proposed system is to detect a variety of routing attacks namely sinkhole, selective forward and blackhole attacks. The detection algorithm uses trust management strategies that are based on a set of trust properties each of which is used for the detection of a specific type of routing attacks. The proposed attack detection algorithm was simulated using the Contiki Cooja simulator with centralized intrusion detection system placement strategy. The evaluation results show that in the proposed algorithm was able to detect the simulated attacks with 100% true positive detection rate in some scenarios.

Increasing confidence for discerning species and population compositions from metabarcoding assays of environmental samples: case studies of fishes in the Laurentian Great Lakes and Wabash River

Community composition data are essential for conservation management, facilitating identification of rare native and invasive species, along with abundant ones. However, traditional capture-based morphological surveys require considerable taxonomic expertise, are time consuming and expensive, can kill rare taxa and damage habitats, and often are prone to false negatives. Alternatively, metabarcoding assays can be used to assess the genetic identity and compositions of entire communities from environmental samples, comprising a more sensitive, less damaging, and relatively time- and cost-efficient approach. However, there is a trade-off between the stringency of bioinformatic filtering needed to remove false positives and the potential for false negatives. The present investigation thus evaluated use of four mitochondrial (mt) DNA metabarcoding assays and a customized bioinformatic Bioinformatic pipeline to increase confidence in species identifications by removing false positives, while achieving high detection probability. Positive controls were used to calculate sequencing error, and results that fell below those cutoff values were removed, unless found with multiple assays. The performance of this approach was tested to discern and identify North American freshwater fishes using lab experiments (mock communities and aquarium experiments) and processing of a bulk ichthyoplankton sample. The method then was applied to field environmental (e) DNA water samples taken concomitant with electrofishing surveys and morphological identifications. This protocol detected 100% of species present in concomitant electrofishing surveys in the Wabash River and an additional 21 that were absent from traditional sampling. Using single 1 L water samples collected from just four locations, the metabarcoding assays discerned 73% of the total fish species that were discerned during four months of an extensive electrofishing river survey in the Maumee River, along with an additional nine species. In both rivers, total fish species diversity was best resolved when all four metabarcoding assays were used together, which identified 35 additional species missed by electrofishing. Ecological distinction and diversity levels among the fish communities also were better resolved with the metabarcoding assays than with morphological sampling and identifications, especially using all four assays together. At the population-level, metabarcoding analyses targeting the invasive round goby Neogobius melanostomus and the silver carp Hypophthalmichthys molitrix identified all population haplotype variants found using Sanger sequencing of morphologically sampled fish, along with additional intra-specific diversity, meriting further investigation. Overall findings demonstrated that the use of multiple metabarcoding assays and custom bioinformatics that filter potential error from true positive detections improves confidence in evaluating biodiversity.

A Generic Approach for Efficient Detection of Vascular Structures

Vascular segmentation is often required in medical image analysis for various imaging modalities. Despite the rich literature in the field, the proposed methods need most of the time adaptation to the particular investigation and may sometimes lack the desired accuracy in terms of true positive and false positive detection rate. This paper proposes a general method for vascular segmentation based on locally connected filtering applied in a multiresolution scheme. The filtering scheme performs progressive detection and removal of the vessels from the image relief at each resolution level, by combining directional 2D-3D locally connected filters (LCF). An important property of the LCF is that it preserves (positive contrasted) structures in the image if they are topologically connected with other similar structures in their local environment. Vessels, which appear as curvilinear structures, can be filtered out by an appropriate LCF set-up which will minimally affect sheet-like structures. The implementation in a multiresolution framework allows dealing with different vessel sizes. The outcome of the proposed approach is illustrated on several image modalities including lung, liver and coronary arteries. It is shown that besides preserving high accuracy in detecting small vessels, the proposed technique is less sensitive with respect to noise and the presence of pathologies of positive-contrast appearance on the images. The detection accuracy is compared with a previously developed approach on the 20 patient database from the VESSEL12 challenge.

A practical method for predicting patient-specific collision in radiotherapy.

PurposeTo develop a practical method for predicting patient‐specific collision during the treatment planning process.Materials and methodBased on geometry information of the accelerator gantry and the location of plan isocenter, the collision‐free space region could be determined. In this study, collision‐free space region was simplified as a cylinder. Radius of cylinder was equal to the distance from isocenter to the collimator cover. The collision‐free space was converted and imported into treatment planning system (TPS) in the form of region of interest (ROI) which was named as ROISS. Collision was viewed and evaluated on the fusion images of patient's CT and ROIs in TPS. If any points of patient's body or couch fell beyond the safety space, collision would occur. This method was implemented in the Pinnacle TPS. The impact of safety margin on accuracy was also discussed. Sixty‐five plans of clinical patients were chosen for the clinical validation.ResultsWhen the angle of couch is zero, the ROISS displays as a series of circles on the cross section of the patient's CT. When the couch angle is not zero, ROISS is a series of ellipses in the transverse view of patient's CT. The ROISS can be generated quickly within five seconds after a single mouse click in TPS. Adding safety margin is an effective measure in preventing collisions from being undetected. Safety margin could increase negative predictive value (NPV) of test cases. Accuracy obtained was 96.3% with the 3 cm safety margin with 100% true positive collision detection.ConclusionThis study provides a reliable, accurate, and fast collision prediction during the treatment planning process. Potential collisions can be discovered and prevented early before delivering. This method can integrate with the current clinical workflow without any additional required resources, and contribute to improvement in the safety and efficiency of the clinic.

Validation prediction: a flexible protocol to increase efficiency of automated acoustic processing for wildlife research.

Automated recognition is increasingly used to extract species detections from audio recordings; however, the time required to manually review each detection can be prohibitive. We developed a flexible protocol called "validation prediction" that uses machine learning to predict whether recognizer detections are true or false positives and can be applied to any recognizer type, ecological application, or analytical approach. Validation prediction uses a predictable relationship between recognizer score and the energy of an acoustic signal but can also incorporate any other ecological or spectral predictors (e.g., time of day, dominant frequency) that will help separate true from false-positive recognizer detections. First, we documented the relationship between recognizer score and the energy of an acoustic signal for two different recognizer algorithm types (hidden Markov models and convolutional neural networks). Next, we demonstrated our protocol using a case study of two species, the Common Nighthawk (Chordeiles minor) and Ovenbird (Seiurus aurocapilla). We reduced the number of detections that required validation by 75.7% and 42.9%, respectively, while retaining at least 98% of the true-positive detections. Validation prediction substantially improves the efficiency of using automated recognition on acoustic data sets. Our method can be of use to wildlife monitoring and research programs and will facilitate using automated recognition to mine bioacoustic data sets.

Early warning signals of population productivity regime shifts in global fisheries

Abstract Early warning signals (EWSs) are indicators of critical transitions, and their ability to forecast ecological regime shifts depends on a priori understanding of underlying mechanisms. Conversely, performances of EWSs may indicate the mechanisms underlying observed ecological regime shifts. I evaluate the performances of two EWSs, i.e. standard deviation (SD) and autoregressive coefficient (AR1), in predicting the observed productivity regime shifts of 191 global fisheries populations. Signals of SD and AR1 are simultaneously detected in less than 50% population regime shifts and SD has higher positive likelihood ratio than AR1, suggesting critical transition may not be the major cause of observed regime shifts. However, when EWSs are correctly detected, they tend to provide more than five years warning time for more than 50% populations. The detection rate and warning time of SD and AR1 differ among fish orders, and rockfish populations show higher probability of true positive detection and longer warning time. EWSs are not definitive indicators of mechanisms underlying ecological regime shifts, but they may provide guidance in situations where ecological knowledge are limited.

Deriving Surface Ages on Mars Using Automated Crater Counting

AbstractImpact craters on solar system bodies are used to determine the relative ages of surfaces. The smaller the limiting primary crater size, the higher the spatial resolution in surface/resurfacing age dating. A manually counted database (Robbins & Hynek, 2012, https://doi.org/10.1029/2011JE003966) of >384,000 craters on Mars >1 km in diameter exists. But because crater size scales as a power law, the number of impact craters in the size range 10 m to 1 km is in the tens of millions, a number making precise analysis of local variations of age, over an entire surface, impossible to perform by manual counting. To decode this crater size population at a planetary scale, we developed an automated Crater Detection Algorithm based on the You Only Look Once v3 object detection system. The algorithm was trained by annotating images of the controlled Thermal Emission Imaging System daytime infrared data set. This training data set contains 7,048 craters that the algorithm used as a learning benchmark. The results were validated against the manually counted database as the ground truth data set. We applied our algorithm to the Thermal Emission Imaging System global mosaic between ±65° of latitude, returning a true positive detection rate of 91% and a diameter estimation error (~15%) consistent with typical manual count variation. Importantly, although a number of automated crater counting algorithms have been published, for the first time we demonstrate that automatic counting can be routinely used to derive robust surface ages.

A Novel Framework for Lung Nodule Classification with Low False Positive and High True Positive Detection Rate

A Novel Framework for Lung Nodule Classification with Low False Positive and High True Positive Detection Rate

Anomaly pattern detection for streaming data

Outlier detection aims to find a data sample that is different from most other data samples. While outlier detection is performed at an individual instance level, anomaly pattern detection on a data stream means detecting a time point where a pattern to generate data is unusual and significantly different from normal behavior. Beyond predicting the outlierness of individual data samples in a data stream, it can be very useful to detect the occurrence of anomalous patterns in real time. In this paper, we propose a method for anomaly pattern detection in a data stream based on binary classification for outliers and statistical tests on a data stream of binary labels of normal or an outlier. In the first step, by applying the clustering-based outlier detection method, we transform a data stream into a stream of binary values where 0 stands for the prediction as normal data and 1 for outlier prediction. In the second step, anomaly pattern detection is performed on a stream of binary values by two approaches: testing the equality of parameters in the binomial distributions of a reference window and a detection window, and using control charts for the fraction defective. The proposed method obtained the average true positive detection rate of 94% in simulated experiments using real and artificial data. The experimental results also show that anomaly pattern occurrence can be detected reliably even when outlier detection performance is relatively low.

Register-Transfer-Level Features for Machine-Learning-Based Hardware Trojan Detection

Register-transfer-level (RTL) information is hardly available for hardware Trojan detection. In this paper, four RTL Trojan features related to branching statement are proposed. The Minimum Redundancy Maximum Relevance (mRMR) feature selection is applied to the proposed Trojan features to determine the recommended feature combinations. The feature combinations are then tested using different machine learning concepts in order to determine the best approach for classifying Trojan and normal branches. The result shows that a Decision Tree classification algorithm with all the four proposed Trojan features can achieve an average true positive detection rate of 93.72% on unseen test data.

Detection of Thyroid Nodules with Ultrasound Images Based on Deep Learning.

Thyroid nodules are a common clinical entity with high incidence. Ultrasound is often employed to detect and evaluate thyroid nodules. The development of an efficient automated method to detect thyroid nodules using ultrasound has the potential to reduce both physician workload and operator-dependence. To study the method of automatic detection of thyroid nodules based on deep learning using ultrasound, and to obtain the detection method with higher accuracy and better performance. A total of 1200 ultrasound images of thyroid nodules and 800 ultrasound thyroid images without nodule are collected. An improved faster R-CNN based detection method of thyroid nodule is proposed. Instead of using VGG16 as the backbone, ResNet is employed as the backbone for faster R-CNN. SVM, CNN and Faster-RCNN methods are used for thyroid nodule detection test. Precision, sensitivity, specificity and F1-score indicators are used to evaluate the detection performance of different methods. The method based on deep learning is superior to that based on SVM. Faster R-CNN method and the improved method are better than CNN method. Compared with VGG16 as the backbone, RestNet101 backbone based faster R-CNN method achieves better thyroid detection effect. From the accuracy index, the proposed method is 0.084, 0.032 and 0.019 higher than SVM, CNN and faster R-CNN, respectively. Similar results can be seen in precision, sensitivity, specificity and F1-Score indicators. The proposed method of deep learning achieves the best performance values with the highest true positive and true negative detection compared to other methods and performs best in the detection of thyroid nodules.

Fireball streak detection with minimal CPU processing requirements for the Desert Fireball Network data processing pipeline

Abstract The detection of fireballs streaks in astronomical imagery can be carried out by a variety of methods. The Desert Fireball Network uses a network of cameras to track and triangulate incoming fireballs to recover meteorites with orbits and to build a fireball orbital dataset. Fireball detection is done on-board camera, but due to the design constraints imposed by remote deployment, the cameras are limited in processing power and time. We describe the processing software used for fireball detection under these constrained circumstances. Two different approaches were compared: (1) A single-layer neural network with 10 hidden units that were trained using manually selected fireballs and (2) a more traditional computational approach based on cascading steps of increasing complexity, whereby computationally simple filters are used to discard uninteresting portions of the images, allowing for more computationally expensive analysis of the remainder. Both approaches allowed a full night’s worth of data (over a thousand 36-megapixel images) to be processed each day using a low-power single-board computer. We distinguish between large (likely meteorite-dropping) fireballs and smaller fainter ones (typical ‘shooting stars’). Traditional processing and neural network algorithms both performed well on large fireballs within an approximately 30 000-image dataset, with a true positive detection rate of 96% and 100%, respectively, but the neural network was significantly more successful at smaller fireballs, with rates of 67% and 82%, respectively. However, this improved success came at a cost of significantly more false positives for the neural network results, and additionally the neural network does not produce precise fireball coordinates within an image (as it classifies). Simple consideration of the network geometry indicates that overall detection rate for triangulated large fireballs is calculated to be better than 99.7% and 99.9%, by ensuring that there are multiple double-station opportunities to detect any one fireball. As such, both algorithms are considered sufficient for meteor-dropping fireball event detection, with some consideration of the acceptable number of false positives compared to sensitivity.

Reference-free transcriptome exploration reveals novel RNAs for prostate cancer diagnosis.

The use of RNA-sequencing technologies held a promise of improved diagnostic tools based on comprehensive transcript sets. However, mining human transcriptome data for disease biomarkers in clinical specimens are restricted by the limited power of conventional reference-based protocols relying on unique and annotated transcripts. Here, we implemented a blind reference-free computational protocol, DE-kupl, to infer yet unreferenced RNA variations from total stranded RNA-sequencing datasets of tissue origin. As a bench test, this protocol was powered for detection of RNA subsequences embedded into putative long noncoding (lnc)RNAs expressed in prostate cancer. Through filtering of 1,179 candidates, we defined 21 lncRNAs that were further validated by NanoString for robust tumor-specific expression in 144 tissue specimens. Predictive modeling yielded a restricted probe panel enabling more than 90% of true-positive detections of cancer in an independent The Cancer Genome Atlas cohort. Remarkably, this clinical signature made of only nine unannotated lncRNAs largely outperformed PCA3, the only used prostate cancer lncRNA biomarker, in detection of high-risk tumors. This modular workflow is highly sensitive and can be applied to any pathology or clinical application.

Earthquake Arrival Association with Backprojection and Graph Theory

AbstractThe association of seismic‐wave arrivals with causative earthquakes becomes progressively more challenging as arrival detection methods become more sensitive, and particularly when earthquake rates are high. For instance, seismic waves arriving across a monitoring network from several sources may overlap in time, false arrivals may be detected, and some arrivals may be of unknown phase (e.g., P or S waves). We propose an automated method to associate arrivals with earthquake sources and obtain source locations applicable to such situations. To do so, we use a pattern detection metric based on the principle of backprojection to reveal candidate sources followed by graph‐theory‐based clustering and an integer linear optimization routine to associate arrivals with the minimum number of sources necessary to explain the data. This method solves for all sources and phase assignments simultaneously, rather than in a sequential greedy procedure as is common in other association routines. We demonstrate our method on both synthetic and real data from the Integrated Plate Boundary Observatory Chile seismic network of northern Chile. For the synthetic tests, we report results for cases with varying complexity, including rates of 500 earthquakes/day and 500 false arrivals/station/day, for which we measure true positive detection accuracy of >95%. For the real data, we develop a new catalog between 1 January 2010 and 31 December 2017 containing 817,548 earthquakes, with detection rates on average 279 earthquakes/day and a magnitude‐of‐completion of M∼1.8. A subset of detections are identified as sources related to quarry and industrial site activity, and we also detect thousands of foreshocks and aftershocks of the 1 April 2014 Mw 8.2 Iquique earthquake. During the highest rate of aftershock activity, >600 earthquakes/day are detected in the vicinity of the Iquique earthquake rupture zone.

Fine-Scale Spatial and Spectral Clustering of UAV-Acquired Digital Aerial Photogrammetric (DAP) Point Clouds for Individual Tree Crown Detection and Segmentation

The field of remote sensing is undergoing rapid changes through the utilization of unmanned aerial vehicle (UAV) technology. The rise of this new technology and the corresponding growth in the application of digital aerial photogrammetric point clouds (DAP) require renewed investigation into individual tree detection (ITD) routines; most of which have been developed for airborne laser scanning (ALS) and traditional aerial imagery. This article analyzes the application of a well-known ALS-ITD routine to UAV-acquired DAP. In particular, specific modifications are proposed and evaluated aimed at improving its applicability to DAP, with particular emphasis on the incorporation of spectral information through subcrown scale k-means clustering. The new routine, which utilizes point-level spectral information in the clustering process, improved overall true positive detection by ∼6.3%, with the most significant improvements in true positive detection found in lower canopy class stems. The new routine was also tested without the inclusion of spectral information and was shown to produce poorer results by ∼4.1%, indicating that the inclusion of these data is beneficial for ITD approaches. This new routine represents an advancement of processing approaches incorporating both structural and spectral components for ITD in the era of UAV remote sensing in forested environments.

P2473Non-invasive assessment of central venous pressure in heart failure: comparing three-dimensional echocardiography and Swan-Ganz catheter

Abstract Background Assessing hemodynamics, in particular central venous pressure (CVP) is essential in heart failure diagnostics, leading individual therapy. Hereby, invasive measurement through Swan-Ganz right heart catheterization (RHC) is considered gold standard for patient evaluation, but catheterization implies risks of invasiveness including bleeding, infection, vessel and nerve injury, as well as patient discomfort. Non-invasive methods are warranted, but no alternative technique is validated yet. Two-dimensional echocardiography (2DE) is believed to be uncertain in this approach as vena cava often shows ellipse-shapes. Therefore, this study sought to investigate standardized and breathing corrected three-dimensional inferior vena cava echocardiography (3DE) to directly compare CVP with right heart catheterization. Methods and results We prospectively included 100 consecutive heart failure patients in this study (mean age 53±12 years, body mass index 27±5, New York Heart Association functional class 2.3±0.6, left ventricular ejection fraction 34.1±12.8%, brain natriuretic peptide 658.13±974.03, 76% male), all underwent Swan-Ganz right-heart catheterization and immediately both 2DE and 3DE (Philips EPIQ 7G) of inferior vena cava. From two-dimensional data the diameter of IVC was measured perpendicularly in long and short-axis. From 3DE data a cross-sectional image of IVC was reconstructed for both vertical and horizontal diameters of IVC as well as the area of IVC. Established 2DE images revealed mean vena cava sizes of 15.9±5.9 mm, while standardized cross-sectional breathing corrected 3DE images showed diameters of 19.8±7.8 mm in longitudinal axis and 15.74±7.8 in short axis. RHC mean CVP was 9.00±5.4 mmHg and correlation of CVP and 2DE measurements failed adequate correlation (2DE 95% CI 0.19–1.61; r=0.25; p=0.312). However, 3DE axis ratio assessment correlated well with invasive CVP and showed reproducible results (3DE 95% CI 0.26–0.69; r=0.89; p<0.01). This resulted for a CVP cut point of 10 mmHg in a 89% true negative and 50% true positive correct detection. Conclusions Standardized 3DE correlates well with invasive CVP while established 2DE usual care assessment does not show reliable CVP correlation. 3DE CVP assessment may represent a more feasible and easily applicable method for CVP measurement, including absence for risks of right heart catheterization. Further studies are ongoing to validate these findings in the future.

Improved Detection of a Mobile Radiological Source by Means of a Temporal Radiation Profile

We demonstrate how the detection efficiency of a single sensor can be improved by means of a temporal radiation count-rate profile in a security checkpoint situation (vehicular or pedestrian). In particular, we show improved true positive detection vis-a-vis a traditional method without increasing the false positive rate. Our results are confirmed by both computer and physical experiments in shielded and unshielded scenarios.