Negative Pixels Research Articles

Improving image-based weld defects detection under interference information via a novel semantic supervision strategy

ABSTRACT Weld defects of skin-skeleton structures are invisible and image-based visual inspections using deep learning neural networks are in demand. The main limitation of previous detection algorithms is interference information in images. The reflective light and uneven brightness distribution hinder algorithms to achieve higher detection reliability. In this work, a novel supervision strategy for defects detection algorithms was proposed to break the limitation. A new semantic gate convolution block was developed to help the neural networks to distinguish between targets and interference. The block utilised semantic segmentation labels and a gate function to control the information transmitted to the detection output. Additionally, a new category-pixel-accuracy loss function was adopted to improve the effectiveness of semantic features. The new function reduced the effects of negative pixels to avoid over emphasis on semantic completeness, thus the ineffective features of interference could be eliminated. The results indicated that the back-propagated gradients from the supervision part taught the backbone networks to focus on correct objects instead of interference. To testify the validity of the current method, the algorithms was verified in burns and collapse defects. The detection accuracy under the new semantic supervision reached 99.1%, which was superior to common CNN.

Forced Measurement of Astronomical Sources at Low Signal-to-noise

We propose a modified moment matching algorithm to avoid catastrophic failures for sources with a low signal to noise ratio. The proposed modifications include a method to eliminate nonphysical negative pixel values and a forced single iteration with an initial guess derived from coadd measurements when iterative methods are unstable. We correct for all biases in measurements introduced by the method. We find that the proposed modifications allow the algorithm to avoid catastrophic failures in nearly 100% of the cases, especially at low SNR. Additionally, with a reasonable guess from coadd measurements, the algorithm measures the flux, centroid, size, shape, and ellipticity with a bias statistically consistent with zero. We show that the proposed method allows us to measure sources 7 times fainter than traditional methods when applied to images obtained from WIYN-ODI. We also present a scheme to find uncertainties in measurements when using the new method to measure astronomical sources.

Functional Optical Coherence Tomography of Rat Cortical Neurovascular Activation during Monopulse Electrical Stimulation with the Microelectrode Array

This paper presents a study to evoke rat cortical functional activities, including hemodynamic and neural tissue signal changes, by monopulse electrical stimulation with a microelectrode array using functional optical coherence tomography (fOCT). Based on the principal component analysis and fuzzy clustering method (PCA-FCM), the hemodynamic response of different size blood vessels in rat cortex are analyzed, showing that the hemodynamic response of the superficial large blood vessels is more concentrated. In the regions of neural tissue where blood vessels are removed, positive significant pixels (the intensity of the pixel for five consecutive frames is greater than the average value plus triple standard deviation) and negative significant pixels (the intensity of the pixel for five consecutive frames is less than the average value minus triple standard deviation) exist, and the averaged intensity signal responds rapidly with an onset time of ~20.8 ms. Furthermore, the hemodynamic response was delayed by ~3.5 s from the neural tissue response. fOCT can provide a label-free, large-scale and depth-resolved map of cortical neurovascular activation, which is a promising technology to monitor cortical small-scale neurovascular activities.

Semi-supervised learning for concrete defect segmentation from images

Deep learning-based concrete defect detection has emerged as a promising technology to overcome the limitations of manual visual inspection, which is often subjective, laborious, and time-consuming. However, most existing methods rely heavily on manual labels and adopt a fully supervised learning approach, which hinders their practical application in engineering. To tackle these concerns, we proposed a semi-supervised learning model, termed Semi-SegDefect. Our proposal utilizes a mean teacher architecture to assign pseudo-labels to the pixels of unlabeled images and performs pixel-level contrastive learning on a sparse set of hard negative pixels to achieve segmentation boundaries with the highest possible accuracy. Test results demonstrate that Semi-SegDefect outperforms fully supervised models significantly, especially when trained with limited labeled data. This method shows great promise for enhancing the accuracy and scalability of concrete defect segmentation and can contribute to the advancement of the construction industry.

Full 3-D modulation transfer function estimation of tomosynthesis system using modified Richardson-Lucy deconvolution.

Tomosynthesis systems are three-dimensional (3-D) medical imaging devices that operate over limited acquisition angles using low radiation dosages. To measure the spatial resolution performance of a tomosynthesis system, the modulation transfer function (MTF) is widely used as a quantitative evaluationmetric. We previously introduced a method to estimate the full 3-D MTF of a cone-beam computed tomography system using two-dimensional (2-D) Richardson-Lucy (RL) deconvolution with Tikhonov-Miller regularization. However, this method can not be applied directly to estimate the 3-D MTF of a tomosynthesis system, since the unique artifacts (i.e., shadow artifacts, spreading tails, directional blurring, and high-level noise) of the system produce several errors that lower the estimation performance. Varying positions of the negative pixels due to shadow artifacts and spreading tails cause inconsistent deconvolution performances at each of the directional projections, and the severe noise in the reconstructed images cause noise amplification during estimation. This work proposes several modifications to the previous method to resolve the inconsistent performance and noise amplification errors to increase the full 3-D MTF estimationaccuracy. Three modifications were introduced to the 2-D RL deconvolution to prevent estimation errors and improve MTF estimation performance: non-negativity relaxation function, cost function to terminate the iterative process of RL deconvolution, and regularization strength for noise control. To validate the effectiveness of the proposed modifications, we reconstructed sphere phantoms from simulation and experimental tomosynthesis studies in the iso-center and offset-center positions as well as estimated the full 3-D MTFs using the previous and proposed methods. We compared the 3-D render images, central plane images, and center profiles of the estimated 3-D MTFs and calculated the full widths at half and tenth maximum for quantitativeevaluation. The previous method cannot estimate the full 3-D MTF of a tomosynthesis system; its inaccurate negative pixel relaxation produces circular-shaped errors, and the mean squared error based simple cost function for termination causes inconsistent estimation at each directional projection to diminish the clear edges of the low-frequency drop and missing sample regions. Noise amplification from lack of noise regularization is also seen in the previous method results. Compared to the previous method, the proposed method shows superior estimation performance at reducing errors in both the simulation and experimental studies regardless of object position. The proposed method preserves the low-frequency drop, missing sample regions from the limited acquisition angles, and missing cone region from the offset-center position; the estimated MTFs also show FWHM and FWTM values close to those of the ideal MTFs than with the previousmethod. This work presents a method to estimate the full 3-D MTF of a tomosynthesis system. The proposed modifications prevent circular-shaped errors and noise amplification due to the geometry for limited acquisition angles and high noise levels. Compared to our previous method, the proposed scheme show better performance for estimating the 3-D MTF of the tomosynthesissystem.

Assessment of vegetation variation and its response to ENSO and IOD in the semi-arid ecosystem of Western India

Abstract The variation in the vegetation pattern reflects the change in the regional environment. Normalized Difference Vegetation Index (NDVI) data from 2000 to 2022 for the Upper Bhima sub-basin in Western India has been used to identify the response of vegetation to the El Niño-Southern Oscillations (ENSO) and Indian Ocean Dipole (IOD) events. As a novelty, the present study identifies the ENSO-sensitive and IOD-sensitive vegetation areas within the watershed using vegetation mean to difference anomalies. Monthly NDVI anomalies are used to determine sensitive pixels of vegetation using mean monthly NDVI. Local spatial autocorrelation (LISA) is performed to analyze the pattern of the NDVI anomalies and cluster maps are generated. The results of spatial variation show that NDVI is adversely affected in El Niño years. During La Niña years, the percentage area covered by dense vegetation is more than 80%, which is significantly higher than that of El Niño years in the monsoon and post-monsoon periods. Positive IOD years show significantly more sparse vegetation cover than negative IOD years. The results of LISA analysis show that the rainfall shadow zone in the study area has a cluster of negative sensitive pixels even in the monsoon and post-monsoon period except in negative IOD year.

BRPPNet: Balanced privacy protection network for referring personal image privacy protection

Traditional personal image privacy protection usually suffers from the overprotection problem, where one or more undesired persons in an image may be inevitably shielded, yielding unnecessary information loss. Motivated by this, this paper explores a novel task “Referring Personal Image Privacy Protection” (RP-IPP) to protect the designated person in an image according to a user’s text or voice input. We propose a lightweight yet effective personal protection network “Balanced Referring Personal PrivacyNet” (BRPPNet), which introduces a Multi-scale Feature Fusion Module (MFFM) with a proper “Balanced-BCE loss” to effectively localize the referring person. Technically, MFFM adopts a lightweight CNN backbone to filter noise information as well as complement visual features for high-quality mask generation. What is more, we have theoretically proven the insufficiency of binary cross-entropy (BCE) loss and its variants for RP-IPP, which suffers from the serious imbalance problem during gradient propagation, and thus formulate “Balanced-BCE loss” to alleviate the gradient propagation bias caused by unequal positive and negative pixels. To verify the effectiveness of BRPPNet, we manually construct a dataset “Referring Personal COCO” (RPCOCO). The experimental results demonstrate that BRPPNet outperforms the advanced approaches for RP-IPP, and the proposed “Balanced-BCE loss” embedded into several existing approaches consistently boosts performance, yielding remarkable improvements on all the metrics.

Automatic Road Extraction with Multi-Source Data Revisited: Completeness, Smoothness and Discrimination

Extracting roads from multi-source data, such as aerial images and vehicle trajectories, is an important way to maintain road networks in the filed of urban computing. In this paper, we revisit the problem of road extraction and aim to boost its accuracy by solving three significant issues: the insufficient complementarity among multiple sources, rough edges of extracted roads, and many false positives caused by confusing pixels. In particular, we design an end-to-end neural network model to achieve this goal. At first, this model leverages two encoding networks to extract relative information from the inputs of two sources respectively, and then applies the attention mechanism to fuse them for sufficiently capturing the complementary correlation. Next, we introduce an auxiliary task, predicting road edges based on fused representations, to make the extracted roads smooth and continuous. At last, to reduce false positives relative to confusing pixels, we propose a pixel-aware contrastive-learning module to distinguish positive (roads) and negative (objects similar to roads) pixels. In addition, to improve the model's learning effectiveness, we propose a model-agnostic transfer learning method, which first builds auxiliary tasks to pre-train the whole model, and then fine-tunes the model's parameters for the main task. Extensive experiments on real datasets verify the superiority of our method as well as the importance of solving the three issues outlined above.

MV-CDN: Multi-Visual Collaborative Deep Network for Change Detection of Double-Temporal Hyperspectral Images

Since individual neural networks have limited deep expressiveness and effectiveness, many learning frameworks face difficulties in the availability and balance of sample selection. As a result, in change detection, it is difficult to upgrade the hit rate of a high-performance model on both positive and negative pixels. Therefore, supposing that the sacrificed components coincide perfectly with the important evaluation objectives, such as positives, it would lose more than gain. To address this issue, in this paper, we propose a multi-visual collaborative deep network (MV-CDN) served by three collaborative network members that consists of three subdivision approaches, the CDN with one collaborator (CDN-C), CDN with two collaborators (CDN-2C), and CDN with three collaborators (CDN-3C). The purpose of the collaborator is to re-evaluate the feature elements in the network transmission, and thus to translate the group-thinking into a more robust field of vision. We use three sets of public double-temporal hyperspectral images taken by the AVIRIS and HYPERION sensors to show the feasibility of the proposed schema. The comparison results have confirmed that our proposed schema outperforms the existing state-of-the-art algorithms on the three tested datasets, which demonstrates the broad adaptability and progressiveness of the proposal.

Coronary X-ray angiography segmentation using Artificial Intelligence: a multicentric validation study of a deep learning model

IntroductionWe previously developed an artificial intelligence (AI) model for automatic coronary angiography (CAG) segmentation, using deep learning. To validate this approach, the model was applied to a new dataset and results are reported.MethodsRetrospective selection of patients undergoing CAG and percutaneous coronary intervention or invasive physiology assessment over a one month period from four centers. A single frame was selected from images containing a lesion with a 50–99% stenosis (visual estimation). Automatic Quantitative Coronary Analysis (QCA) was performed with a validated software. Images were then segmented by the AI model. Lesion diameters, area overlap [based on true positive (TP) and true negative (TN) pixels] and a global segmentation score (GSS – 0 -100 points) - previously developed and published - were measured.Results123 regions of interest from 117 images across 90 patients were included. There were no significant differences between lesion diameter, percentage diameter stenosis and distal border diameter between the original/segmented images. There was a statistically significant albeit minor difference [0,19 mm (0,09–0,28)] regarding proximal border diameter. Overlap accuracy ((TP + TN)/(TP + TN + FP + FN)), sensitivity (TP / (TP + FN)) and Dice Score (2TP / (2TP + FN + FP)) between original/segmented images was 99,9%, 95,1% and 94,8%, respectively. The GSS was 92 (87–96), similar to the previously obtained value in the training dataset.Conclusionthe AI model was capable of accurate CAG segmentation across multiple performance metrics, when applied to a multicentric validation dataset. This paves the way for future research on its clinical uses.

A novel chaotic DICOM image encryption algorithm based on pseudo-random polyline and rapid diffusion-scrambling

This paper proposes a novel chaotic DICOM image encryption algorithm based on pseudo-random polyline and rapid diffusion-scrambling. This paper proposes to preprocess the image twice in the encryption process to ensure that any type of image can be encrypted, including images with negative pixel values, which increases the practicability of the algorithm. Two pseudo-random polylines are generated to segment the image during the encryption process. The pseudo-random polylines are generated from the information of the plaintext image, which ensures that each image produces different pseudo-random polylines and increases the security of the algorithm. In the encryption process, rapid diffusion-scrambling is used to avoid the waste of extra space. And piecewise diffusion and non-adjacent matrix reading are adopted to make the encryption more unpredictable. The scheme is analyzed from the performance indicators such as pixel correlation and key space, and the results demonstrated strong security and performance of the algorithm.

Bootstrapping Semi-supervised Medical Image Segmentation with Anatomical-Aware Contrastive Distillation.

Contrastive learning has shown great promise over annotation scarcity problems in the context of medical image segmentation. Existing approaches typically assume a balanced class distribution for both labeled and unlabeled medical images. However, medical image data in reality is commonly imbalanced (i.e., multi-class label imbalance), which naturally yields blurry contours and usually incorrectly labels rare objects. Moreover, it remains unclear whether all negative samples are equally negative. In this work, we present ACTION, an Anatomical-aware ConTrastive dIstillatiON framework, for semi-supervised medical image segmentation. Specifically, we first develop an iterative contrastive distillation algorithm by softly labeling the negatives rather than binary supervision between positive and negative pairs. We also capture more semantically similar features from the randomly chosen negative set compared to the positives to enforce the diversity of the sampled data. Second, we raise a more important question: Can we really handle imbalanced samples to yield better performance? Hence, the key innovation in ACTION is to learn global semantic relationship across the entire dataset and local anatomical features among the neighbouring pixels with minimal additional memory footprint. During the training, we introduce anatomical contrast by actively sampling a sparse set of hard negative pixels, which can generate smoother segmentation boundaries and more accurate predictions. Extensive experiments across two benchmark datasets and different unlabeled settings show that ACTION significantly outperforms the current state-of-the-art semi-supervised methods.

Differential diagnostic value of plain CT scan in adrenal adenoma and non-adenoma: A two-center control study of mean attenuation value, minimum attenuation value, and CT histogram.

To investigate the value of mean attenuation value (AVmean), minimum attenuation value (AVmin), and CT histogram (CTH) for the differential diagnosis of adrenal adenoma and non-adenoma in two medical centers. The plain CT data of 403 cases of adrenal adenoma and 141 cases of non-adenoma in center A were retrospectively analyzed, and compared with data of 86 cases of adenoma and 71 cases of non-adenoma in center B. All cases were confirmed by pathology or clinical follow-up. The diagnostic efficacy of AVmean ≤ 10 Hounsfield units (HU), AVmin ≤ 0 HU, and CTH negative pixels ≥ 10% for adrenal adenoma, and AVmin and CTH for adenoma with AVmean > 10Hu were compared between the two medical centers. In medical centers A and B, the AUC of AVmean for the differential diagnosis of adenoma and non-adenoma was 0.956 and 0.956, respectively, and the corresponding sensitivity, specificity, and accuracy were, 0.591 and 0.663, 1.000 and 1.000, 0.697, and 0.815, respectively, when the threshold was ≤ 10 HU. The AUC of AVmin was 0.941 and 0.958, respectively, and the corresponding sensitivity, specificity, and accuracy were 0.869 and 0.826, 0.986, and 0.972, 0.899, and 0.892, respectively, when the threshold was ≤ 0 HU. The AUC of CTH negative pixels was 0.948 and 0.952, respectively, and the corresponding sensitivity, specificity, and accuracy were 0.759 and 0.674, 1.000 and 1.000, 0.822, and 0.822, respectively, when the threshold was ≥ 10%. Among adenoma with AVmean >10 HU, the best threshold of AVmin in center A and center B were -0.250HU and 2.375HU, and the corresponding AUC, sensitivity and specificity were 0.858 and 0.846, 0.691 and 0.586, 0.986 and 0.958; the best threshold of CTH in center A and center B were 0.895% and 0.775%, and the corresponding AUC, sensitivity and specificity were 0.873 and 0.822, 0.818 and 0.724, 0.837 and 0.915. AVmean, AVmin, and CTH are all important parameters for differentiating adrenal adenoma from non-adenoma. Even for adenomas with AVmean > 10 HU, AVmin and CTH still had high diagnostic efficiency. The three parameters are complementary, assisting clinicians to develop personalized treatments.

Evaluation of near-infrared hyperspectral imaging for the assessment of potato processing aptitude.

The potato (Solanum tuberosum L.) is the world's fifth most important staple food with high socioeconomic relevance. Several potato cultivars obtained by selection and crossbreeding are currently on the market. This diversity causes tubers to exhibit different behaviors depending on the processing to which they are subjected. Therefore, it is interesting to identify cultivars with specific characteristics that best suit consumer preferences. In this work, we present a method to classify potatoes according to their cooking or frying as crisps aptitude using NIR hyperspectral imaging (HIS) combined with a Partial Least Squares Discriminant Analysis (PLS-DA). Two classification approaches were used in this study. First, a classification model using the mean spectra of a dataset composed of 80 tubers belonging to 10 different cultivars. Then, a pixel-wise classification using all the pixels of each sample of a small subset of samples comprised of 30 tubers. Hyperspectral images were acquired using fresh-cut potato slices as sample material placed on a mobile platform of a hyperspectral system in the NIR range from 900 to 1,700 nm. After image processing, PLS-DA models were built using different pre-processing combinations. Excellent accuracy rates were obtained for the models developed using the mean spectra of all samples with 90% of tubers correctly classified in the external dataset. Pixel-wise classification models achieved lower accuracy rates between 66.62 and 71.97% in the external validation datasets. Moreover, a forward interval PLS (iPLS) method was used to build pixel-wise PLS-DA models reaching accuracies above 80 and 71% in cross-validation and external validation datasets, respectively. Best classification result was obtained using a subset of 100 wavelengths (20 intervals) with 71.86% of pixels correctly classified in the validation dataset. Classification maps were generated showing that false negative pixels were mainly located at the edges of the fresh-cut slices while false positive were principally distributed at the central pith, which has singular characteristics.

Segmentation and volume quantification of epicardial adipose tissue in computed tomography images.

Many cardiovascular diseases are closely related to the composition of epicardial adipose tissue (EAT). Accurate segmentation of EAT can provide a reliable reference for doctors to diagnose the disease. The distribution and composition of EAT often have significant individual differences, and the traditional segmentation methods are not effective. In recent years, deep learning method has been gradually introduced into EAT segmentation task. The existing EAT segmentation methods based on deep learning have a large amount of computation and the segmentation accuracy needs to be improved. Therefore, the purpose of this paper is to develop a lightweight EAT segmentation network, which can obtain higher segmentation accuracy with less computation and further alleviate the problem of false-positive segmentation. First, the obtained computed tomography was preprocessed. That is, the threshold range of EAT was determined to be -190, -30HU according to prior knowledge, and the non-adipose pixels were excluded by threshold segmentation to reduce the difficulty of training. Second, the image obtained after thresholding was input into the lightweight RDU-Net network to perform the training, validating, and testing process. RDU-Net uses a residual multi-scale dilated convolution block in order to extract a wider range of information without changing the current resolution. At the same time, the form of residual connection is adopted to avoid the problem of gradient expansion or gradient explosion caused by too deep network, which also makes the learning easier. In order to optimize the training process, this paper proposes PNDiceLoss, which takes both positive and negative pixels as learning targets, fully considers the class imbalance problem, and appropriately highlights the status of positive pixels. In this paper, 50 CCTA images were randomly selected from the hospital, and the commonly used Dice similarity coefficient (DSC), Jaccard similarity, accuracy (ACC), specificity (SP), precision (PC), and Pearson correlation coefficient are used as evaluation metrics. Bland-Altman analysis results show that the extracted EAT volume is consistent with the actual volume. Compared with the existing methods, the segmentation results show that the proposed method achieves better performance on these metrics, achieving the DSC of 0.9262. The number of false-positive pixels has been reduced by more than half. Pearson correlation coefficient reached 0.992, and linear regression coefficient reached 0.977 when measuring the volume of EAT obtained. In order to verify the effectiveness of the proposed method, experiments are carried out in the cardiac fat database of VisualLab. On this database, the proposed method also achieved good results, and the DSC value reached 0.927 in the case of only 878 slices. A new method to segment and quantify EAT is proposed. Comprehensive experiments show that compared with some classical segmentation algorithms, the proposed method has the advantages of shorter time-consuming, less memory required for operations, and higher segmentation accuracy. The code is available at https://github.com/lvanlee/EAT_Seg/tree/main/EAT_seg.

Active Fire Detection from Landsat-8 Imagery Using Deep Multiple Kernel Learning

Active fires are devastating natural disasters that cause socio-economical damage across the globe. The detection and mapping of these disasters require efficient tools, scientific methods, and reliable observations. Satellite images have been widely used for active fire detection (AFD) during the past years due to their nearly global coverage. However, accurate AFD and mapping in satellite imagery is still a challenging task in the remote sensing community, which mainly uses traditional methods. Deep learning (DL) methods have recently yielded outstanding results in remote sensing applications. Nevertheless, less attention has been given to them for AFD in satellite imagery. This study presented a deep convolutional neural network (CNN) “MultiScale-Net” for AFD in Landsat-8 datasets at the pixel level. The proposed network had two main characteristics: (1) several convolution kernels with multiple sizes, and (2) dilated convolution layers (DCLs) with various dilation rates. Moreover, this paper suggested an innovative Active Fire Index (AFI) for AFD. AFI was added to the network inputs consisting of the SWIR2, SWIR1, and Blue bands to improve the performance of the MultiScale-Net. In an ablation analysis, three different scenarios were designed for multi-size kernels, dilation rates, and input variables individually, resulting in 27 distinct models. The quantitative results indicated that the model with AFI-SWIR2-SWIR1-Blue as the input variables, using multiple kernels of sizes 3 × 3, 5 × 5, and 7 × 7 simultaneously, and a dilation rate of 2, achieved the highest F1-score and IoU of 91.62% and 84.54%, respectively. Stacking AFI with the three Landsat-8 bands led to fewer false negative (FN) pixels. Furthermore, our qualitative assessment revealed that these models could detect single fire pixels detached from the large fire zones by taking advantage of multi-size kernels. Overall, the MultiScale-Net met expectations in detecting fires of varying sizes and shapes over challenging test samples.

PixelGame: Infrared Small Target Segmentation as a Nash Equilibrium

A key challenge of infrared small target segmentation (ISTS) is to balance false negative pixels (FNs) and false positive pixels (FPs). Traditional methods combine FNs and FPs into a single objective by weighted sum, and the optimization process is decided by one actor. Minimizing FNs and FPs with the same strategy leads to antagonistic decisions. To address this problem, we propose a competitive game framework (pixelGame) from a novel perspective for ISTS. In pixelGame, FNs and FPs are controlled by different player whose goal is to minimize their own utility function. FNs-player and FPs-player are designed with different strategies: One is to minimize FNs and the other is to minimize FPs. The utility function drives the evolution of the two participants in competition. We consider the Nash equilibrium of pixelGame as the optimal solution. In addition, we propose maximum information modulation (MIM) to highlight the target information. MIM effectively focuses on the salient region including small targets. Extensive experiments on two standard public datasets prove the effectiveness of our method. Compared with other state-of-the-art methods, our method achieves better performance in terms of F1-measure ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathrm{F_{1}}$</tex-math></inline-formula> ) and the intersection of union ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{IoU}$</tex-math></inline-formula> ).

Unenhanced Dual-Layer Spectral-Detector CT for Characterizing Indeterminate Adrenal Lesions.

Background Unenhanced dual-layer spectral-detector CT may facilitate adrenal lesion characterization; however, no studies have evaluated its incremental diagnostic yield for indeterminate lesions (unenhanced attenuation >10 HU) in comparison to that with conventional unenhanced CT. Purpose To determine whether spectral attenuation analysis improves characterization of lipid-poor adrenal adenomas from nonadenomas compared to that with mean attenuation and histogram analysis of conventional CT images. Materials and Methods This retrospective study included patients with indeterminate adrenal lesions who underwent unenhanced dual-layer spectral-detector CT between March 2018 and June 2020. Mean attenuation on conventional 120-kVp images (HUconv), histogram-based percentage negative pixels (proportion of all pixels <0 HU) on conventional 120-kVp images, and mean attenuation on virtual monoenergetic images (VMIs) at 40-140 keV were measured for each lesion. The attenuation difference between virtual monoenergetic 140- and 40-keV images (ΔHU; ie, Hounsfield unit at 140 keV - Hounsfield unit at 40 keV) and ΔHU indexed with HUconv (ΔHU index; ie, ΔHU/HUconv × 100) were calculated. Conventional and virtual monoenergetic imaging parameters were compared between lipid-poor adenomas and nonadenomas by using the Mann-Whitney U test. Receiver operating characteristic analysis was performed to determine the sensitivity for attaining at least 95% specificity in characterizing adenomas from nonadenomas; sensitivity was compared by using the McNemar test. Results A total of 232 patients (mean age ± standard deviation, 67 years ± 11; 145 men) with 129 lipid-poor adenomas and 103 nonadenomas were evaluated. HUconv and mean attenuation on VMIs at 40-140 keV were lower and the percentage negative pixels, ΔHU, and ΔHU index higher in lipid-poor adenomas than in nonadenomas (P < .001 for all). Attenuation differences between adenomas and nonadenomas on VMIs were maximal at 40 keV (23 HU at 40 keV vs 5 HU at 140 keV). The highest sensitivities for differentiating adenomas and nonadenomas were achieved for virtual monoenergetic ΔHU index (77% [99 of 129 adenomas]), attenuation on 40-keV images (71% [91 of 129 adenomas]), and ΔHU (67% [87 of 129 adenomas]) compared to HUconv (35% [45 of 129 adenomas]) and percentage negative pixels (30% [39 of 129 adenomas]) (P < .001 for all; specificity, 95% [98 of 103 nonadenomas]). Conclusion Spectral attenuation analysis enabled differentiation of lipid-poor adenomas from nonadenomas with higher sensitivity than mean attenuation or histogram analysis of conventional CT images. © RSNA, 2021 Online supplemental material is available for this article.

Reducing metal artifacts by restricting negative pixels

When the object contains metals, its x-ray computed tomography (CT) images are normally affected by streaking artifacts. These artifacts are mainly caused by the x-ray beam hardening effects, which deviate the measurements from their true values. One interesting observation of the metal artifacts is that certain regions of the metal artifacts often appear as negative pixel values. Our novel idea in this paper is to set up an objective function that restricts the negative pixel values in the image. We must point out that the naïve idea of setting the negative pixel values in the reconstructed image to zero does not give the same result. This paper proposes an iterative algorithm to optimize this objective function, and the unknowns are the metal affected projections. Once the metal affected projections are estimated, the filtered backprojection algorithm is used to reconstruct the final image. This paper applies the proposed algorithm to some airport bag CT scans. The bags all contain unknown metallic objects. The metal artifacts are effectively reduced by the proposed algorithm.

Self-Supervised Saliency Estimation for Pixel Embedding in Road Detection

Road detection is an important task inthe signal processing field. Although self-supervised learning has the potential to learn rich and effective visual representations that avoid tedious labeling, the current approaches learn from object-centered images, which leads to ambiguous results in complex traffic scenarios. We introduce saliency estimation to extend the self-supervised segmentation beyond object-center images, with spatial-temporal information and ensemble learning employed to improve the robustness. Then, we also design a quadruple loss for the pixel embedding learning and optimize the affinity between different categories, while exploring structural information in negative pixels. Experiments on the public datasets show that our approach is competitive with state-of-the-art approaches.