False Positive Suppression Research Articles

Transductive meta-learning with enhanced feature ensemble for few-shot semantic segmentation

This paper addresses few-shot semantic segmentation and proposes a novel transductive end-to-end method that overcomes three key problems affecting performance. First, we present a novel ensemble of visual features learned from pretrained classification and semantic segmentation networks with the same architecture. Our approach leverages the varying discriminative power of these networks, resulting in rich and diverse visual features that are more informative than a pretrained classification backbone that is not optimized for dense pixel-wise classification tasks used in most state-of-the-art methods. Secondly, the pretrained semantic segmentation network serves as a base class extractor, which effectively mitigates false positives that occur during inference time and are caused by base objects other than the object of interest. Thirdly, a two-step segmentation approach using transductive meta-learning is presented to address the episodes with poor similarity between the support and query images. The proposed transductive meta-learning method addresses the prediction by first learning the relationship between labeled and unlabeled data points with matching support foreground to query features (intra-class similarity) and then applying this knowledge to predict on the unlabeled query image (intra-object similarity), which simultaneously learns propagation and false positive suppression. To evaluate our method, we performed experiments on benchmark datasets, and the results demonstrate significant improvement with minimal trainable parameters of 2.98M. Specifically, using Resnet-101, we achieve state-of-the-art performance for both 1-shot and 5-shot Pascal-5i\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$5^{i}$$\\end{document}, as well as for 1-shot and 5-shot COCO-20i\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$20^{i}$$\\end{document}.

Pulmonary Nodule Detection from 3D CT Image with a Two-Stage Network

Early detection of lung nodules is an important means of reducing the lung cancer mortality rate. In this paper, we propose a three-dimensional CT image lung nodule detection method based on parallel pooling and dense blocks, which includes two parts, i.e., candidate nodule extraction and false positive suppression. First, a dense U-shaped backbone network with parallel pooling is proposed to obtain the candidate nodule probability map. The parallel pooling structure uses multiple pooling operations for downsampling to capture spatial information comprehensively and address the problem of information loss resulting from maximum and average pooling in the shallow layers. Then, a parasitic network with parallel pooling, dense blocks, and attention modules is designed to suppress false positive nodules. The parasitic network takes the multiscale feature maps of the backbone network as the input. The experimental results demonstrate that the proposed method significantly improves the accuracy of lung nodule detection, achieving a CPM score of 0.91, which outperforms many existing methods.

Exploration of motion inhibition for the suppression of false positives in biologically inspired small target detection algorithms from a moving platform

Detecting small moving targets against a cluttered background in visual data is a challenging task. The main problems include spatio-temporal target contrast enhancement, background suppression and accurate target segmentation. When targets are at great distances from a non-stationary camera, the difficulty of these challenges increases. In such cases the moving camera can introduce large spatial changes between frames which may cause issues in temporal algorithms; furthermore targets can approach a single pixel, thereby affecting spatial methods. Previous literature has shown that biologically inspired methods, based on the vision systems of insects, are robust to such conditions. It has also been shown that the use of divisive optic-flow inhibition with these methods enhances the detectability of small targets. However, the location within the visual pathway the inhibition should be applied was ambiguous. In this paper, we investigated the tunings of some of the optic-flow filters and use of a nonlinear transform on the optic-flow signal to modify motion responses for the purpose of suppressing false positives and enhancing small target detection. Additionally, we looked at multiple locations within the biologically inspired vision (BIV) algorithm where inhibition could further enhance detection performance, and look at driving the nonlinear transform with a global motion estimate. To get a better understanding of how the BIV algorithm performs, we compared to other state-of-the-art target detection algorithms, and look at how their performance can be enhanced with the optic-flow inhibition. Our explicit use of the nonlinear inhibition allows for the incorporation of a wider dynamic range of inhibiting signals, along with spatio-temporal filter refinement, which further increases target-background discrimination in the presence of camera motion. Extensive experiments shows that our proposed approach achieves an improvement of 25% over linearly conditioned inhibition schemes and 2.33 times the detection performance of the BIV model without inhibition. Moreover, our approach achieves between 10 and 104 times better detection performance compared to any conventional state-of-the-art moving object detection algorithm applied to the same, highly cluttered and moving scenes. Applying the nonlinear inhibition to other algorithms showed that their performance can be increased by up to 22 times. These findings show that the application of optic-flow- based signal suppression should be applied to enhance target detection from moving platforms. Furthermore, they indicate where best to look for evidence of such signals within the insect brain.

Ion-shift reagent binding energy and the mass-mobility shift correlation in ion mobility spectrometry.

Ion mobility spectrometry (IMS) detects illegal substances and explosives in airports, ports, and customs. This is complicated by false positives caused by overlapping peaks. Shift reagents selectively change ion mobilities through adduction with analyte ions. This discriminates false positives because interferents and illegal substances respond differently to shift reagents. We introduced five different shift reagents using electrospray ionization-IMS-mass spectrometry to study the effect of interaction energy, intermolecular bonds, and analyte size on ion mobility shifts. Analyte ion-shift reagent interactions were calculated using Gaussian. The mobility shifts showed a decreasing trend as the molecular weight increased for a series of ten compounds. The shift in drift time better reflected the pure effect of shift reagents. Valinol was an exception to this trend because it had a low binding energy interaction with all shift reagents and, consequently, its clusters were short-lived. This produced fewer collisions against the buffer gas and a shorter drift time, compared to ions of similar molecular weight. The results of this investigation are important for understanding the behavior of shift reagents in resolving overlapping peaks that cause interferences. The suppression of false positives eases the transit of passengers and cargos, increases the confiscation of illicit substances, and saves money and distress due to needless delays in customs and airports.

Pixel-Level Analysis for Enhancing Threat Detection in Large-Scale X-ray Security Images

Threat detection in X-ray security images is critical for preserving public safety. Recently, deep learning algorithms have begun to be adopted for threat detection tasks in X-ray security images. However, most of the prior works in this field have largely focused on using image-level classification and object-level detection approaches. Adopting object separation as a pixel-level approach to analyze X-ray security images can significantly improve automatic threat detection. In this paper, we investigated the effects of incorporating segmentation deep learning models in the threat detection pipeline of a large-scale imbalanced X-ray dataset. We trained a Faster R-CNN (region-based convolutional neural network) model to localize possible threat regions in the X-ray security images on a balanced dataset to maximize detection of true positives. Then, we trained a DeepLabV3+ model to verify the preliminary detections by classifying each pixel in the threat regions, which resulted in the suppression of false positives. The two models were combined in one detection pipeline to produce the final detections. Experiment results demonstrate that the proposed method significantly outperformed previous baseline methods and end-to-end instance segmentation methods, achieving mean average precisions (mAPs) of 94.88%, 91.40%, and 89.42% across increasing scales of imbalance in the practical dataset.

Automatic detection on intracranial aneurysm from digital subtraction angiography with cascade convolutional neural networks

BackgroundAn intracranial aneurysm is a cerebrovascular disorder that can result in various diseases. Clinically, diagnosis of an intracranial aneurysm utilizes digital subtraction angiography (DSA) modality as gold standard. The existing automatic computer-aided diagnosis (CAD) research studies with DSA modality were based on classical digital image processing (DIP) methods. However, the classical feature extraction methods were badly hampered by complex vascular distribution, and the sliding window methods were time-consuming during searching and feature extraction. Therefore, developing an accurate and efficient CAD method to detect intracranial aneurysms on DSA images is a meaningful task.MethodsIn this study, we proposed a two-stage convolutional neural network (CNN) architecture to automatically detect intracranial aneurysms on 2D-DSA images. In region localization stage (RLS), our detection system can locate a specific region to reduce the interference of the other regions. Then, in aneurysm detection stage (ADS), the detector could combine the information of frontal and lateral angiographic view to identify intracranial aneurysms, with a false-positive suppression algorithm.ResultsOur study was experimented on posterior communicating artery (PCoA) region of internal carotid artery (ICA). The data set contained 241 subjects for model training, and 40 prospectively collected subjects for testing. Compared with the classical DIP method which had an accuracy of 62.5% and an area under curve (AUC) of 0.69, the proposed architecture could achieve accuracy of 93.5% and the AUC of 0.942. In addition, the detection time cost of our method was about 0.569 s, which was one hundred times faster than the classical DIP method of 62.546 s.ConclusionThe results illustrated that our proposed two-stage CNN-based architecture was more accurate and faster compared with the existing research studies of classical DIP methods. Overall, our study is a demonstration that it is feasible to assist physicians to detect intracranial aneurysm on DSA images using CNN.

Detection of atrial fibrillation using discrete-state Markov models and Random Forests

In this paper, we present a fully automated technique for robust detection of Atrial Fibrillation (AF) episodes in single-lead electrocardiogram (ECG) signals using discrete-state Markov models and Random Forests. Methods: The ECG signal is first preprocessed using Stationary Wavelet Transforms (SWT) for noise suppression, signal quality assessment and subsequent R-peak detection. Discrete-state Markov probabilities modelling transitions between successive RR intervals along with other statistical quantities derived from the RR-interval series constitute the feature set to perform AF classification using Random Forests. Further enhancement in AF detection is achieved by using a post-processing false positive suppression algorithm based on autocorrelation analysis of the RR-interval series. Datasets: The AF classifier was trained using the Physionet/Computing in Cardiology 2017 AF Challenge dataset and the Atrial Fibrillation Termination Database (AFTDB). The test datasets consist of the MIT-BIH Atrial Fibrillation Database (AFDB) and the MIT-BIH Arrhythmia Database (MITDB). Results: Our algorithms achieved sensitivity, specificity and F-score values of 97.4%, 98.6% and 97.7% respectively on the AFDB dataset and 96.3%, 97.0% and 85.6% respectively on the MITDB dataset. It was also observed that inclusion of the false positive suppression step resulted in a 1.1% increase in specificity and a 4.0% increase in F-score for the MITDB dataset without any decrease in sensitivity. Conclusion: The proposed method of AF detection, combining Markov models and Random Forests, achieves high accuracy across multiple databases and demonstrates comparable or superior performance to several other state-of-the-art algorithms.

Accurate Detection of Inner Ears in Head CTs Using a Deep Volume-to-Volume Regression Network with False Positive Suppression and a Shape-Based Constraint.

Cochlear implants (CIs) are neural prosthetics which are used to treat patients with hearing loss. CIs use an array of electrodes which are surgically inserted into the cochlea to stimulate the auditory nerve endings. After surgery, CIs need to be programmed. Studies have shown that the spatial relationship between the intra-cochlear anatomy and electrodes derived from medical images can guide CI programming and lead to significant improvement in hearing outcomes. However, clinical head CT images are usually obtained from scanners of different brands with different protocols. The field of view thus varies greatly and visual inspection is needed to document their content prior to applying algorithms for electrode localization and intra-cochlear anatomy segmentation. In this work, to determine the presence/absence of inner ears and to accurately localize them in head CTs, we use a volume-to-volume convolutional neural network which can be trained end-to-end to map a raw CT volume to probability maps which indicate inner ear positions. We incorporate a false positive suppression strategy in training and apply a shape-based constraint. We achieve a labeling accuracy of 98.59% and a localization error of 2.45 mm. The localization error is significantly smaller than a random forest-based approach that has been proposed recently to perform the same task.

Indexing for local appearance-based recognition of planar objects

Local appearance-based approaches for planar object recognition do not support efficient means for indexing of object models. The size of the feature set which needs to be stored and processed is also large. To reduce database size, we propose an optimal feature extraction technique that selects only the salient features of an object. Since in typical local appearance-based systems, the actual feature information is not isolated from the background, scene clutter causes error in recognition. We propose a mechanism whereby this shortcoming can be alleviated. Further, by indexing onto this space, we propose to improve the performance in terms of computation time and suppression of false positives. This indexing mechanism based on the color histogram and the geometry of the feature, exploits the fact that features tend to form clusters in the feature space based on their similarity of appearances. We have implemented the proposed system and verified its validity through extensive testing.

Multidimensional indexing for recognizing visual shapes

This paper introduces an analytical framework for studying some properties of model acquisition and recognition techniques based on indexing. The goal is to demonstrate that several problems previously associated with the approach can be attributed to the low dimensionality of invariants used. These include limited index selectivity. Excessive accumulation of votes in the look-up table buckets, and excessive sensitivity to quantization parameters. Theoretical results demonstrate that using high-dimensional, highly descriptive global invariants produces better results in terms of accuracy, false positive suppression, and computation time. A practical example of high-dimensional global invariants is introduced and used to implement a 2-D shape acquisition/recognition system. The acquisition/recognition system is based on a two-step table look-up mechanism. First, local curve descriptors are obtained by correlating image contour information at short range. Then, seven-dimensional global invariants are computed by correlating triplets of local curve descriptors at longer range. This experimental system is meant to illustrate the behavior of a high-dimensional indexing scheme. Indeed, its performance shows good agreement with the analytical model with respect to database size, fault tolerance, and recognition speed. Model acquisition time is linear to cubic in the number of object features. Object recognition time is constant to linear in the number of models in the database and linear to cubic in the number of features in the image. The system has been tested extensively. With more than 250 arbitrary shapes in the database. Unsupervised shape and subpart acquisition is demonstrated.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>