Dynamically Spatiotemporal Regularized Correlation Tracking

Learning Self-Corrective Network via Adaptive Self-Labeling and Dynamic NMS for High-Performance Long-Term Tracking

This article presents a self-corrective network-based long-term tracker (SCLT) including a self-modulated tracking reliability evaluator (STRE) and a self-adjusting proposal postprocessor (SPPP). The targets in the long-term sequences often suffer from severe appearance variations. Existing long-term trackers often online update their models to adapt the variations, but the inaccurate tracking results introduce cumulative error into the updated model that may cause severe drift issue. To this end, a robust long-term tracker should have the self-corrective capability that can judge whether the tracking result is reliable or not, and then it is able to recapture the target when severe drift happens caused by serious challenges (e.g., full occlusion and out-of-view). To address the first issue, the STRE designs an effective tracking reliability classifier that is built on a modulation subnetwork. The classifier is trained using the samples with pseudo labels generated by an adaptive self-labeling strategy. The adaptive self-labeling can automatically label the hard negative samples that are often neglected in existing trackers according to the statistical characteristics of target state, and the network modulation mechanism can guide the backbone network to learn more discriminative features without extra training data. To address the second issue, after the STRE has been triggered, the SPPP follows it with a dynamic NMS to recapture the target in time and accurately. In addition, the STRE and the SPPP demonstrate good transportability ability, and their performance is improved when combined with multiple baselines. Compared to the commonly used greedy NMS, the proposed dynamic NMS leverages an adaptive strategy to effectively handle the different conditions of in view and out of view, thereby being able to select the most probable object box that is essential to accurately online update the basic tracker. Extensive evaluations on four large-scale and challenging benchmark datasets including VOT2021LT, OxUvALT, TLP, and LaSOT demonstrate superiority of the proposed SCLT to a variety of state-of-the-art long-term trackers in terms of all measures. Source codes and demos can be found at https://github.com/TJUT-CV/SCLT.

Active Learning for Deep Visual Tracking

Convolutional neural networks (CNNs) have been successfully applied to the single target tracking task in recent years. Generally, training a deep CNN model requires numerous labeled training samples, and the number and quality of these samples directly affect the representational capability of the trained model. However, this approach is restrictive in practice, because manually labeling such a large number of training samples is time-consuming and prohibitively expensive. In this article, we propose an active learning method for deep visual tracking, which selects and annotates the unlabeled samples to train the deep CNN model. Under the guidance of active learning, the tracker based on the trained deep CNN model can achieve competitive tracking performance while reducing the labeling cost. More specifically, to ensure the diversity of selected samples, we propose an active learning method based on multiframe collaboration to select those training samples that should be and need to be annotated. Meanwhile, considering the representativeness of these selected samples, we adopt a nearest-neighbor discrimination method based on the average nearest-neighbor distance to screen isolated samples and low-quality samples. Therefore, the training samples' subset selected based on our method requires only a given budget to maintain the diversity and representativeness of the entire sample set. Furthermore, we adopt a Tversky loss to improve the bounding box estimation of our tracker, which can ensure that the tracker achieves more accurate target states. Extensive experimental results confirm that our active-learning-based tracker (ALT) achieves competitive tracking accuracy and speed compared with state-of-the-art trackers on the seven most challenging evaluation benchmarks. Project website: https://sites.google.com/view/altrack/.

Studying the Factors of Human Carotid Atherosclerotic Plaque Rupture, by Calculating Stress/Strain in the Plaque, Based on CEUS Images: A Numerical Study

Carotid plaque neovascularization is one of the major factors for the classification of vulnerable plaque, but the axial force effects of the pulsatile blood flow on the plaque with neovessel and intraplaque hemorrhage was unclear. Together with the severity of stenosis, the fibrous cap thickness, large lipid core, and the neovascularization followed by intraplaque hemorrhage (IPH) have been regarded as high-risk features of plaque rupture. In this work, the effects of these factors were evaluated on the progression and rupture of the carotid atherosclerotic plaques. Five geometries of carotid artery plaque were developed based on contrast-enhanced ultrasound (CEUS) images, which contain two types of neovessel and IPH, and geometry without neovessel and IPH. A one-way fluid-structure interaction model was applied to compute the maximum principal stress and strain in the plaque. For that hyper-elastic and non-linear material, Yeoh 3rd Order strain energy density function was used for components of the plaque. The simulation results indicated that the maximum principal stress of plaque in the carotid artery was higher when the degree of the luminal stenosis increased and the thickness of the fibrous cap decreased. The neovessels within the plaque could introduce a 2.5% increments of deformation in the plaque under the pulsatile blood flow pressure. The IPH also contributed to the increased risk of plaque rupture that a gain of stress was 8.983, 14.526, and 34.47 kPa for the plaque with 50, 65, and 75%, respectively, when comparing stress in the plaque with IPH distributed at the middle to the shoulder of the plaque. In conclusion, neovascularization in the plaque could reduce the stability of the plaque by increasing the stress within the plaque. Also, the risk of plaque rupture increased when large luminal stenosis, thin fibrous cap, and IPH were observed.

Non-invasive Evaluation of Brain Death Caused by Traumatic Brain Injury by Ultrasound Imaging

Objectives

To investigate the clinical value of non-invasive ultrasound imaging in the evaluation of brain death caused by traumatic brain injury.

Methods

Thirty-four patients with acute severe traumatic brain injury were admitted to hospital within 48 h after injury. All patients were monitored intracranial pressure, transcranial Doppler, echocardiography examination, collection intracranial pressure, MCA-Vs, MCA-Vd, MCA-Vm, EF, LVMPI, RVMPI and other indicators, and combined with clinical conditions and other related data for comparative study and statistical analysis.

Results

The blood flow spectrum was characterized by diastolic retrograde blood flow spectrum pattern and nail waveform spectrum shape when the patient had clinical brain death. For the parameters of transcranial Doppler, there were significant differences in MCA-Vm and PI between clinical brain death group and normal control group (P < 0.05). For the parameters of echocardiography, there were statistically significant differences in EF, LVMPI, and RVMPI between clinical brain death group and normal control group (P < 0.05).

Conclusion

Non-invasive dynamic monitoring of cerebral hemodynamics and cardiac function parameters in patients with severe craniocerebral injury can provide a high accuracy and reliability for the preliminary diagnosis of brain death in patients with severe craniocerebral injury. It is helpful for early evaluation of prognosis and provides effective monitoring methods and guidance for clinical treatment.

Review of Deep Learning Approaches for the Segmentation of Multiple Sclerosis Lesions on Brain MRI

In recent years, there have been multiple works of literature reviewing methods for automatically segmenting multiple sclerosis (MS) lesions. However, there is no literature systematically and individually review deep learning-based MS lesion segmentation methods. Although the previous review also included methods based on deep learning, there are some methods based on deep learning that they did not review. In addition, their review of deep learning methods did not go deep into the specific categories of Convolutional Neural Network (CNN). They only reviewed these methods in a generalized form, such as supervision strategy, input data handling strategy, etc. This paper presents a systematic review of the literature in automated multiple sclerosis lesion segmentation based on deep learning. Algorithms based on deep learning reviewed are classified into two categories through their CNN style, and their strengths and weaknesses will also be given through our investigation and analysis. We give a quantitative comparison of the methods reviewed through two metrics: Dice Similarity Coefficient (DSC) and Positive Predictive Value (PPV). Finally, the future direction of the application of deep learning in MS lesion segmentation will be discussed.

Classification of Schizophrenia by Functional Connectivity Strength Using Functional Near Infrared Spectroscopy

Functional near-infrared spectroscopy (fNIRS) has been widely employed in the objective diagnosis of patients with schizophrenia during a verbal fluency task (VFT). Most of the available methods depended on the time-domain features extracted from the data of single or multiple channels. The present study proposed an alternative method based on the functional connectivity strength (FCS) derived from an individual channel. The data measured 100 patients with schizophrenia and 100 healthy controls, who were used to train the classifiers and to evaluate their performance. Different classifiers were evaluated, and support machine vector achieved the best performance. In order to reduce the dimensional complexity of the feature domain, principal component analysis (PCA) was applied. The classification results by using an individual channel, a combination of several channels, and 52 ensemble channels with and without the dimensional reduced technique were compared. It provided a new approach to identify schizophrenia, improving the objective diagnosis of this mental disorder. FCS from three channels on the medial prefrontal and left ventrolateral prefrontal cortices rendered accuracy as high as 84.67%, sensitivity at 92.00%, and specificity at 70%. The neurophysiological significance of the change at these regions was consistence with the major syndromes of schizophrenia.

DFENet: Deep Feature Enhancement Network for Accurate Calculation of Instantaneous Wave-Free Ratio

Accurate iFR calculation can provide important clinical information for intracoronary functional assessment without administration of adenosine, which needs to locate object points in the pressure waveforms: peak, the dichrotic notch and the pressure nadir at the end of diastole. We propose a DFENet that is capable of locating object points to calculate iFR accurately. We first design a SFRA into DFENet with the idea of DenseNet. To avoid overfitting when dealing with sparse signals, we set appropriate number of network layers, growth rate of dense blocks and compression rate of transition blocks in 1D DenseNet. Then, we introduce a feature enhancement mechanism named 1D SE block for enhancing inconspicuous but vital features from SFRA, which guides DFENet to focus on these important features via feature recalibration. Finally, we prove an effective interaction mode between SFRA and 1D SE block to locate object points accurately. Adequate experiments demonstrate that DFENet reaches a high accuracy of 94.22%, error of 5.6 on object point localization of 1D pressure waveforms that include 1457 samples from 100 subjects via a cross-validation of Leave-One-Out. Comparison experiment demonstrates that the accuracy of DFENet exceeds other state-of-the-art methods by 3.35%, and ablation experiment demonstrates that the accuracy of SFRA and cSE exceed the other variations by 6.63% and 2.56% respectively. Importantly, we reveal how the DFENet enhance inconspicuous but vital feature by applying gradient-weighted class activation maps. DFENet can locate object points accurately, which is applicable to other signal processing tasks, especially in health sensing.