Image Sequences Research Articles

Predicting high-risk group according to Oncotype DX recurrence score using dynamic contrast-enhanced breast MR with temporal radiomic features.

e12533 Background: Temporal radiomic features (TRF) extracted from dynamic contrast-enhanced breast MR (DCE-MR), which provide important information about tumor heterogeneity, offer a non-invasive approach to predict high-risk groups and could potentially serve as a cost-effective alternative to the genetic assay. This study explored how TRF from breast MR could be integrated to predict the high-risk group of OncotypeDX (ODX). Methods: In 173 patients with breast cancer [low-risk, 144 (83.2%); high-risk, 29 (16.8%)], TRF such as dynamic signal intensity changes and texture variations were derived from the imaging sequences. Hierarchical clustering was applied to reduce feature redundancy and identify significant predictors. Machine learning algorithms such as random forest, SVM, logistic regression, and KNN were utilized with 7-fold cross validation model. Results: Cross-validation revealed that models with TRF were consistently better than those without it across 4 different machine learning algorithms. Random forest AUC improved from 0.48 to 0.56, support vector machines from 0.46 to 0.63, logistic regression from 0.51 to 0.69, and K-nearest neighbors from 0.61 to 0.73 (Table). Conclusions: Incorporating TRF from DCE-MR images into a machine learning model improved the predictive performance for high-risk groups of ODX compared to using only the reference RF. Comparison of classifier performance using radiomic and temporal radiomic features. Random forest SVM Logistic regression KNN Features RF RF + TRF RF RF + TRF RF RF + TRF RF TRF AUC Fold1 0.38 (0.13, 0.65) 0.41 (0.14, 0.69) 0.68 (0.38, 0.94) 0.54 (0.23, 0.84) 0.39 (0.12, 0.71) 0.50 (0.18, 0.85) 0.60 (0.22, 0.90) 0.55 (0.27, 0.82) AUC Fold2 0.62 (0.35, 0.86) 0.52 (0.24, 0.85) 0.71 (0.45, 0.92) 0.65 (0.40, 0.87) 0.31 (0.09, 0.57) 0.55 (0.28, 0.84) 0.54 (0.29, 0.78) 0.69 (0.42, 0.91) AUC Fold3 0.57 (0.30, 0.82) 0.68 (0.38, 0.93) 0.46 (0.10, 0.75) 0.50 (0.12, 0.86) 0.49 (0.13, 0.83) 0.76 (0.53, 0.96) 0.43 (0.00, 0.82) 0.70 (0.48, 0.90) AUC Fold4 1.00 (1.00, 1.00) 0.83 (0.67, 0.96) 0.08 (0.00, 0.21) 0.79 (0.62, 0.96) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 0.77 (0.61, 0.90) 0.75 (0.56, 0.92) AUC Fold5 0.23 (0.00, 0.50) 0.30 (0.00, 0.63) 0.83 (0.62, 0.98) 0.44 (0.13, 0.83) 0.36 (0.08, 0.62) 0.44 (0.13, 0.82) 0.62 (0.27, 0.89) 0.66 (0.46, 0.83) AUC Fold6 0.23 (0.00, 0.50) 0.66 (0.39, 0.89) 0.35 (0.00, 0.78) 0.76 (0.57, 0.95) 0.70 (0.44, 0.91) 0.70 (0.37, 1.00) 0.62 (0.13, 0.95) 0.85 (0.62, 1.00) AUC Fold7 0.35 (0.08, 0.67) 0.48 (0.14, 0.91) 0.11 (0.00, 0.32) 0.75 (0.53, 0.91) 0.33 (0.08, 0.67) 0.86 (0.68, 1.00) 0.70 (0.45, 0.90) 0.90 (0.77, 1.00) AUC Average 0.48 0.56 0.46 0.63 0.51 0.69 0.61 0.73 Data were expressed as Area Under the Curve (AUC) and 95% Confidence Intervals (CI). RF, Radiomic features; TRF, Temporal radiomic features; SVM, Support Vector Machine; KNN, K-Nearest Neighbors, AUC, Area Under the ROC Curve.

A novel hybrid generation technique of facial expressions using fine-tuning and auxiliary condition generative adversarial network

The facial expression generation continues to be interesting for researchers and scientists worldwide. Facial expressions are an excellent way of transmitting one's emotions or intentions to others, and they have been extensively studied in areas such as driver safety, human-computer interaction (HCI), deception detection, health care, and monitoring. The facial expression generation starts with a single neutral image and generates sequences of facial expression images, which are combined to create a video. Previous methods generated facial expression images of the same person. However, they still suffer from low accuracy and image quality. This article overcomes this problem using a novel hybrid model for facial expression video generation using fine-tuning and condition-generative model architectures to optimize the model's parameters. Results indicate that the proposed novel approach significantly improves the expression generation of the same person. The proposed method can reliably and accurately generate facial expressions, with a testing accuracy of 98.7% and a training accuracy of 99.9%.

Optimizing knee joint imaging: A comparative study of 7T MRI sequences.

Optimizing knee joint imaging: A comparative study of 7T MRI sequences.

Assessment of hemodynamic disturbances and impaired ventricular filling in asymptomatic fontan patients: A 4D flow CMR study.

Assessment of hemodynamic disturbances and impaired ventricular filling in asymptomatic fontan patients: A 4D flow CMR study.

Accuracy of magnetic resonance imaging black bone sequence for mandibular fracture diagnosis and dislocation measurement.

Accuracy of magnetic resonance imaging black bone sequence for mandibular fracture diagnosis and dislocation measurement.

NaMA-Mamba: Foundation model for generalizable nasal disease detection using masked autoencoder with Mamba on endoscopic images.

NaMA-Mamba: Foundation model for generalizable nasal disease detection using masked autoencoder with Mamba on endoscopic images.

New joint estimation method for emissivity and temperature distribution based on a Kriged Marginalized Particle Filter: Application to simulated infrared thermal image sequences

New joint estimation method for emissivity and temperature distribution based on a Kriged Marginalized Particle Filter: Application to simulated infrared thermal image sequences

Self-supervised combustion state diagnosis using a noise-augmented generative adversarial network and flame image sequences

Self-supervised combustion state diagnosis using a noise-augmented generative adversarial network and flame image sequences

Delivering Gd-Labeled IgG Antibodies Into the Mouse Brain Following Focused Ultrasound Treatment.

Antibody-based therapy has emerged as a powerful tool for targeted treatment of neurological diseases, such as brain cancer and neurodegenerative disorders. However, direct, scalable, and safe confirmation of antibody delivery into the brain remains challenging. Antibodies can be effectively tracked when tagged with molecules that are detectable by medical imaging modalities, such as MRI, PET, or SPECT. In this study, we aimed to confirm gadolinium (Gd)-labeled IgG antibody delivery into the mouse brain using MRI, following exposure to focused ultrasound (FUS) and circulating microbubbles. We acquired MR images of the mouse brain to evaluate antibody delivery into the targeted brain region. First, we quantified the MR signal of Gd-labeled IgG antibodies in phantoms using preclinical 9.4 T and clinical 3 T MRI scanners. Then, we determined optimal ultrasound and MR imaging parameters to non-invasively and safely disrupt the blood-brain barrier in a localized and reversible manner and effectively monitor antibody delivery into the murine brain, respectively. We confirmed that IgG antibodies can be reliably delivered into the murine brain using FUS and microbubble treatment and that we can track their biodistribution within the brain parenchyma using clinically relevant MR image sequences. The maximum detected volume of Gd-IgG antibody delivery (n = 4) was determined to be 0.12 ± 0.02 mm3 at t = 75.3 ± 17.3 minutes following treatment. This work paves the way for a scalable and non-ionizing method for performing and evaluating antibody delivery into the brain.

DWCLF-Net: A weighted contrastive learning feature fusion network for temporal scar image sequence classification

DWCLF-Net: A weighted contrastive learning feature fusion network for temporal scar image sequence classification

A Lightweight Pedestrian Tracking Target Detection Algorithm

In recent years, pedestrian detection and tracking has been widely used in the fields of multimedia, computer vision and pattern recognition. Most of the traditional pedestrian tracking algorithms directly use the pixel level features in each frame of the image sequence in the video to calculate, ignoring the deep semantic feature information contained in each image, resulting in inaccurate or wrong positioning. Moreover, the traditional algorithm has a large number of parameters, high complexity and poor real-time performance. This paper proposes a new lightweight object detection algorithm Res-YOLO-tiny, which aims to accelerate the tracking efficiency of blurred image removal algorithms. It greatly improves the detection speed, and then combines with the PID (proportion integration differentiation) algorithm to increase the tracking effect and positioning accuracy of the whole tracking system through a new ROI (region of interest) combination method. Finally, the experimental and practical results show that the proposed algorithm has high positioning accuracy, sensitive tracking response, strong recognition robustness and flexibility.

DenLsNet-C: a novel model for breast cancer classification in pathology images based on DenseNet and LSTM

In the contemporary world, breast cancer is a common malignancy, whose early detection and timely treatment can increase the patients’ survival prospects. The automated classification of breast cancer types based on histopathological images is a challenging endeavor, wherein computer-assisted diagnosis serves as a reference for pathologists’ decision-making. Addressing the automated breast cancer classification task, this paper proposes a novel DenLsNet neural network model, featuring a combined DenseNet−LSTM architecture for efficient feature extraction and classification. First, the feature extraction process is optimized by incorporating squeeze-and-excitation (SE) blocks into a pretrained improved dense convolutional network (DenseNet). Next, iterative convolutional feature fusion (iCFF) blocks are introduced for deep and shallow feature fusion. To enhance the classification performance, the original DenseNet classifier is replaced with a specially designed long short-term memory (LSTM)-based classifier, which proves effective in capturing long-distance relationships in image sequences, improving the model’s sensitivity to breast cancer variations. Performance evaluation experiments, conducted on the BreakHis and BACH public datasets, demonstrate significant performance enhancement in the multi-class classification task, with DenLsNet exhibiting superior performance compared to state-of-the-art models. Additionally, the proposed model achieves commendable results in the binary classification task, indicating strong generalization capabilities.

Open-ended Autoregressive Visual Storytelling via Parameter Efficient Instruction Tuning

Visual storytelling (VIST) involves generating coherent, creative, and vivid narrative for a collection of images. It remains an immense challenge within cross-modal domain. Traditional mainstream storytelling work were less proficient in handling long sequential relationships. Though large-scale visual-language pre-training (VLP) models demonstrated promising prospect on cross-modal tasks. So far they still were not particularly adept at handling tasks involving image sequences. Moreover, the reference descriptions in the available VIST benchmark dataset are short and simplistic, which constrains model’s potential capabilities. Current models struggle to produce truly rich and vivid narratives. Therefore, in this paper, we will address these deficiencies, and contribute from both dataset and innovative model aspects. Firstly, by leveraging large language model (LLM), we have constructed a new dataset VIST++ which can enrich vivid narratives for open-ended image sequences. The dataset has potential on providing beneficial support for future model learning. Secondly, we have proposed an innovative auto-regressive story generation model named ReStoryGen. It can be applied to image sequences of varying lengths in an open-ended way. We have performed extensive experiments and evaluations in terms of visual grounding, coherence and non-redundancy. The experiment results have convincingly demonstrated ReStoryGen achieves impressive outcomes through utilizing parameter-efficient instruction-tuning. Related source codes and models are distributed on Github https://github.com/lixinliu1995/story_gen.

Assessing the Feasibility of a Spatio-Temporal Approach for Recognizing Microorganisms in Image Sequences

ABSTRACTData classification is one of the main challenges in data mining andknowledge discovery. It is widely explored in multiple applications,such as identifying microorganisms by analyzing images of culturesgrown in Petri dishes. This work proposes to chronologically organizeimages of bacteria on solid media, captured at equidistant intervalsover several days, treating them as video frames to improve therecognition of these microorganisms. To this end, we develop an approachthat chains classification models, integrating spatial featuresfrom pre-trained convolutional neural networks with temporal informationpropagated through target meta-features, i.e., classifieroutputs.We experimentally compared our proposal with two intentionallydesigned baseline methods using a dataset with 240 images,48 per class, and considering the macro-averaged F1 score. Resultsdemonstrate that addressing chronological relations enhances theidentification models’ performance, even though baseline strategiesmay benefit from fewer examples.

Comparison of DSO and ORB‐SLAM3 in Low‐Light Environments With Auxiliary Lighting and Deep Learning Based Image Enhancing

ABSTRACTIn the evolving landscape of robotic navigation, the demand for solutions capable of operating in challenging scenarios, such as low‐light environments, is increasing. This study investigates the performance of two state‐of‐the‐art (SOTA) visual simultaneous localization and mapping (VSLAM) algorithms, direct sparse odometry (DSO) and ORBSLAM3, in their monocular implementation, in the dark indoor scenarios where the only light source is provided by an auxiliary light system installed on a robot. A modified Pioneer3‐DX robot, equipped with a monocular camera, LED bars, and a lux meter, is utilized to collect a novel data set, “LUCID—Lighting Up Campus Indoor Spaces Data Set,” in real‐world, low‐light indoor environments. The data set includes image sequences enhanced using a generative adversarial network (GAN) to simulate varying levels of image enhancement. Through comprehensive experiments, we assess the performances of the V‐SLAM algorithm, considering the critical balance between maintaining adequate auxiliary illumination and enhancing. This study provides insights into the optimization of robotic navigation in lowlight conditions, paving the way for more robust and reliable autonomous navigation systems.

Delayed cerebral infarction in pneumococcal meningitis due to postinfectious vasculitis: A case report.

Delayed cerebral infarction (DCI) is a rare but devastating complication of bacterial meningitis that is characterized by an initial good recovery followed by sudden deterioration several days after presentation. The potential mechanisms underlying DCI encompass postinfectious vasculitis, vasospasm, and cerebral thrombosis. Sequential magnetic resonance angiography and characteristic high-resolution magnetic resonance patterns are helpful in discerning the cause of DCI. A right-handed 45-year-old man with nasal leakage of cerebrospinal fluid from a car accident was admitted to the local hospital for fever, vomiting, and impaired consciousness. Post-infectious vasculitis. Meropenem (1 g 3 times/day), vancomycin (0.5 g 4 times/day), and dexamethasone (10 mg/day, gradually reduced to withdraw) were administered immediately after the patient was diagnosed with Streptococcus pneumoniae meningitis. After the patient was diagnosed with DCI, we applied nimodipine (10 mg 3 times/day) to prevent vasospasm, clopidogrel (75 mg/day), and heparin to counteract thrombosis, and prednisone (60 mg/day) was applied to reduce the inflammatory reaction. On the 52nd day, the patient's magnetic resonance angiography reexamination indicated a continuous expansion of the cerebral infarction area and a worsening of cerebrovascular occlusion. After the prednisone dose was increased, the consciousness disorder ceased to progress. Early detection and treatment are very important for DCI, the sequential magnetic resonance imaging and high-resolution magnetic resonance imaging are conducive to identify the etiology of DCI and providing guidance for treatment. The delayed tapering of glucocorticoids appears to be beneficial for alleviating the severity of DCI.

"Innumerable" lesion burden on brain MRI ‒ a diagnostic approach.

The MR imaging lesion burden described as "innumerable" is rare, and can present a diagnostic challenge. Brain MR imaging with this descriptive term has not been systematically studied. We determine risk factors and MR imaging sequences helpful in a diagnostic algorithm for innumerable brain MR lesions. Twelve thousand four hundred ninety-five brain MR imaging studies done at our institution from July 1, 2013 to June 30, 2016 were surveyed for the term "innumerable". Inclusion criteria included 50 or more parenchymal lesions. Patients were classified into active and chronic groups, based on MR characteristics and clinical features. One hundred and twenty three reports contained the term "innumerable". Thirty-one met inclusion criteria and 19 showed active, and 12 chronic brain process. The active group included 9 metastasis, 6 infarction, 2 microbleeds, and one each, foreign body granulomatous reaction and fungal abscesses. The MR feature accompanied or heralded onset of illness in eight patients. Malignancy was a risk factor in nine patients with metastasis and in 2 with infarct from cancer-associated hypercoagulation. Other risk factors included immunosuppression, endocarditis, long bone fracture and aortic dissection. MR defined active innumerable brain lesions occurred in 0.25 % of studies. Fifteen of 19 in the active group were due to metastasis or infarction, defined by T1-weighted gadolinium enhancement and restricted-diffusion respectively. A diagnostic algorithm based on MR imaging features and risk factors can guide critical decision for brain biopsy.

Investigating the Role of Intravoxel Incoherent Motion Diffusion-Weighted Imaging in Evaluating Multiple Sclerosis Lesions.

Background: Multiple sclerosis (MS) is a chronic and heterogeneous disease characterized by demyelination and axonal loss and damage. Magnetic resonance imaging (MRI) has been employed to distinguish these changes in various types of MS lesions. Objectives: We aimed to evaluate intravoxel incoherent motion (IVIM) diffusion and perfusion MRI metrics across different brain regions in healthy individuals and various types of MS lesions, including enhanced, non-enhanced, and black hole lesions. Methods: A prospective study included 237 patients with MS (65 males and 172 females) and 29 healthy control participants (25 males and 4 females). The field strength was 1.5 Tesla. The imaging sequences included three-dimensional (3D) T1, 3D fluid-attenuated inversion recovery, two-dimensional (2D) T1, T2-weighted imaging, and 2D diffusion-weighted imaging (DWI) sequences. IVIM-derived parameters-apparent diffusion coefficient (ADC), pure molecular diffusion (D), pseudo-diffusion (D*), and perfusion fraction (f)-were quantified for commonly observed lesion types (2506 lesions from 224 patients with MS, excluding 13 patients due to MRI artifacts or not meeting the diagnostic criteria for RR-MS) and for corresponding brain regions in 29 healthy control participants. A one-way analysis of variance, followed by post-hoc analysis (Tukey's test), was performed to compare mean values between the healthy and MS groups. Receiver operating characteristic curve analyses, including area under the curve, sensitivity, and specificity, were conducted to determine the cutoff values of IVIM parameters for distinguishing between the groups. A p-value of ≤0.05 and 95% confidence intervals were used to report statistical significance and precision, respectively. Results: All IVIM parametric maps in this study discriminated among most MS lesion types. ADC, D, and D* values for MS black hole lesions were significantly higher (p < 0.0001) than those for other MS lesions and healthy controls. ADC, D, and D* maps demonstrated high sensitivity and specificity, whereas f maps exhibited low sensitivity but high specificity. Conclusions: IVIM parameters provide valuable diagnostic and clinical insights by demonstrating high sensitivity and specificity in evaluating different categories of MS lesions.

Novel Aortic Dissection Model Links Endothelial Dysfunction and Immune Infiltration.

Aortic dissection (AD) is the separation of medial layers of the aorta and is a major cause of death in patients with connective tissue disorders such as Marfan syndrome. However, molecular triggers instigating AD, its temporospatial progression, and how vascular cells in each vessel layer interact and participate in the pathological process remain incompletely understood. To unravel the underlying molecular mechanisms of AD, we generated a spontaneous AD mouse model. We incorporated a novel missense variant (p.G234D) in FBN1, the gene for fibrillin-1, identified in a patient with nonsyndromic familial AD into mice using the CRISPR/Cas9 system. We performed molecular pathological analyses of the aortic lesions by histology, immunofluorescence staining, electron microscopy, synchrotron-based imaging, and single-cell RNA sequencing. Biochemical analysis was performed to examine the binding capacity of mutant human FBN1G234D protein to LTBPs (latent TGFβ [transforming growth factor-beta] binding proteins), and signaling pathways in the mutant aortic wall were examined by the Western blot analysis. Fifty percent of the Fbn1G234D/G234D mutant mice died within 5 weeks of age from multiple intimomedial tears that expanded longitudinally and progressed to aortic rupture accompanied by massive immune cell infiltration. Fbn1G234D/G234D endothelial cells exhibited altered mechanosensing with loss of parallel alignment to blood flow and upregulation of VCAM-1 and ICAM-1 as early as 1 week of age. Single-cell RNA sequencing, validated by immunostaining, revealed a cluster of monocyte/macrophage predominantly in the intima at 3 weeks of age before the dissection, and the second cluster of macrophages increased during the progression of intimomedial tears, exhibiting strong CCR2+ and both M1- and M2-like features. Consistently, upregulation of MMP2/9 was observed. Biochemically, FBN1G234D lost the ability to bind to LTBP-1, -2, and -4, resulting in the downregulation of TGFβ signaling in the aortic wall. We show that interactions involving endothelial cells and macrophages/monocytes in the intima, where the ECM (extracellular matrix) microenvironment contains reduced TGFβ signaling, contribute to the initiation of AD. Our novel AD mouse model provides a unique opportunity to identify target molecules involved in the intimomedial tears that can be utilized for the development of therapeutic strategies.

A Camera Calibration Method for Temperature Measurements of Incandescent Objects Based on Quantum Efficiency Estimation.

High-temperature thermal images enable monitoring and controlling processes in metal, semiconductors, and ceramic manufacturing but also monitor activities of volcanoes or contrasting wildfires. Infrared thermal cameras require knowledge of the emissivity coefficient, while multispectral pyrometers provide fast and accurate temperature measurements with limited spatial resolution. Bayer-pattern cameras offer a compromise by capturing multiple spectral bands with high spatial resolution. However, temperature estimation from color remains challenging due to spectral overlaps among the color filters in the Bayer pattern, and a widely accepted calibration method is still missing. In this paper, the quantum efficiency of an imaging system including the camera sensor, lens, and filters is inferred from a sequence of images acquired by looking at a black body source between 700 °C and 1100 °C. The physical model of the camera, based on the Planck law and the optimized quantum efficiency, allows the calculation of the Planckian locus in the color space of the camera. A regression neural network, trained on a synthetic dataset representing the Planckian locus, predicts temperature pixel by pixel in the 700 °C to 3500 °C range from live images. Experiments done with a color camera, a multispectral camera, and a furnace for heat treatment of metals as ground truth show that our calibration procedure leads to temperature prediction with accuracy and precision of a few tens of Celsius degrees in the calibration temperature range. Tests on a temperature-calibrated halogen bulb prove good generalization capability to a wider temperature range while being robust to noise.