Search

Background: Benign Paroxysmal Positional Vertigo (BPPV) is the leading cause of vertigo, and its characteristic nystagmus induced by positional maneuvers makes it a good model for Artificial Intelligence (AI) diagnosis. However, during the testing procedure, up to 10 min of indivisible long-range temporal correlation data are produced, making the AI-informed real-time diagnosing unlikely in clinical practice., Methods: A combined 1D and Deep-Learning (DL) composite model was proposed. Two separate cohorts were recruited, with one for model generation and the other for evaluation of model's real-world generalizability. Eight features, including two head traces and three eye traces and their corresponding slow phase velocity (SPV) value, were served as the inputs. Three candidate models were tested, and a sensitivity study was conducted to determine the saliently important features., Results: The study included 2671 patients in the training cohort and 703 in the test cohort. A hybrid DL model achieved a micro-area under the receiver operating curve (AUROC) of 0.982 (95% CI 0.965, 0.994) and macro-AUROC of 0.965 (95% CI 0.898, 0.999) for overall classification. The highest accuracy was observed for right posterior BPPV, with an AUROC of 0.991 (95% CI 0.972, 1.000), followed by left posterior BPPV, with an AUROC of 0.979 (95% CI 0.940, 0.998), the lowest AUROC was 0.928 (95% CI 0.878, 0.966) for lateral BPPV. The SPV was consistently identified as the most predictive feature in the models. If the model process is carried out 100 times for a 10-min data, one single running takes 0.79 ± 0.06 s., Conclusion: This study designed DL models which can accurately detect and categorize the subtype of BPPV, enabling a quick and straightforward diagnosis of BPPV in clinical setting. The critical feature identified in the model helps expand our understanding of this disorder., (© 2023. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany.)

Background and Objective: High-dimensional data generally contains more accurate information for medical image, e.g., computerized tomography (CT) data can depict the three dimensional structure of organs more precisely. However, the data in high-dimension often needs enormous computation and has high memory requirements in the deep learning convolution networks, while dimensional reduction usually leads to performance degradation., Methods: In this paper, a two-dimensional deep learning segmentation network was proposed for medical volume data based on multi-pinacoidal plane fusion to cover more information under the control of computation.This approach has conducive compatibility while using the model proposed to extract the global information between different inputs layers., Results: Our approach has worked in different backbone network. Using the approach, DeepUnet's Dice coefficient (Dice) and Positive Predictive Value (PPV) are 0.883 and 0.982 showing the satisfied progress. Various backbones can enjoy the profit of the method., Conclusions: Through the comparison of different backbones, it can be found that the proposed network with multi-pinacoidal plane fusion can achieve better results both quantitively and qualitatively., (Copyright © 2021 Elsevier B.V. All rights reserved.)

Aiming at the problems in text-to-image such as single local texture, unclear edge details and non-conformity to the input text description, RAGAN is proposed as a text-to-image method based on attention model with increased gate mechanism. To address the problem that the traditional method cannot generate fine-grained images, an attention model network with an added gate mechanism is used to filter out relevant word vectors and combine them with intermediate hidden vectors to form new hidden vectors, and then the mutual game of the generative adversarial network allows the generator to generate images with richer textures and clearer edges of the target objects, thus improving the image quality. To address the problem that the generated images do not match the input text description, text reconstruction is used to extract the deep semantic features embedded in the generated images and compare them with the semantic features of the input text to improve the semantic consistency by defining the reconstruction loss. Compared to the baseline model, the Inception Score and R-precision on the CUB dataset improved by 9.17% and 8.3% respectively, and the Inception Score and R-precision on the COCO dataset improved by 13.67% and 5.56% respectively, demonstrating that the model in this paper is effective in improving the authenticity and artistry of the generated images while maintaining semantic consistency. [ABSTRACT FROM AUTHOR]

Depth estimation from a camera is an important task for 3D perception. Recently, without using the labeled ground truth of depth map, a self-supervised deep learning network can use relative pose to synthesize the target image from the reference image, and the photometric error between synthesized reference image and real one is used as self-supervisory signal. In this paper, we propose a novel self-supervised depth estimation network, which takes advantage of the quadtree constraint to optimize the depth estimation network. Based on the quadtree constraint, the photometric loss and depth loss of quadtree are proposed. In order to solve the problem that multiple depth values in repeated structures and uniform texture regions can cause relatively low photometric loss, we use quadtree-based photometric loss, which calculates the averaged photometric loss in quadtree blocks instead of the pixel-wise loss. For the problem of imbalanced depth distribution, we use quadtree depth loss, which constrains the depth inconsistency within quadtree blocks. The depth estimation network is composed of deep fusion module and cross-layer feature fusion module, which can better extract the feature information of RGB image and sparse keypoints depths, and makes full use of the detail information of the shallow feature map and the semantic information of the deep feature map to enrich the feature information extraction. Experimental results demonstrate that our method outperforms the state-of-the-art approaches of depth estimation. [ABSTRACT FROM AUTHOR]

The automatic segmentation of coronary artery in coronary computed tomography angiography (CCTA) image is of great significance for clinicians to evaluate patients with coronary heart disease. When a 3D image is limited by the amount of available GPU memory, reducing the resolution of 3D image will easily lead to the loss of image detail information. Taking patches of image as input cannot make full use of image context information. Image segmentation based on deep learning is difficult to recover perfect smooth edges. The use of smooth loss function may filter out some small lesions on the coronary artery. In this paper, we present a novel CCTA image segmentation framework that combines deep learning and digital image processing algorithms to address these challenging problems. We first use V-Net to process the CCTA image with lower resolution, and get the basic feature map (rough segmentation result) with the same resolution as the original CCTA image. Then, the original CCTA image is concatenated to the basic feature map and input it into the patch-based cascaded V-shaped module to obtain a accurate coronary artery segmentation image. Finally, the center points of coronary segmentation image and the basic gradient image of the original coronary image are obtained by morphological operation. The center points of coronary artery segmentation image are used as seed points, region growing is performed on the binary basic gradient image until the white contour boundary is searched, so as to obtain a coronary segmentation result with full segmentation and smooth edges. The proposed method is analyzed quantitatively and qualitatively, and the results show that the method is better than the mainstream baseline. The ablation experiment also proved the effectiveness of each module. [ABSTRACT FROM AUTHOR]

Purpose: Medical image segmentation is an essential component of medical image analysis. Accurate segmentation can assist doctors in diagnosis and relieve their fatigue. Although several image segmentation methods based on U‐Net have been proposed, their performances have been observed to be suboptimal in the case of small‐sized objects. To address this shortcoming, a novel network architecture is proposed in this study to enhance segmentation performance on small medical targets. Methods: In this paper, we propose a joint multi‐scale context attention network architecture to simultaneously capture higher level semantic information and spatial information. In order to obtain a greater number of feature maps during decoding, the network concatenates the images of side inputs by down‐sampling during the encoding phase. In the bottleneck layer of the network, dense atrous convolution (DAC) and multi‐scale residual pyramid pooling (RMP) modules are exploited to better capture high‐level semantic information and spatial information. To improve the segmentation performance on small targets, the attention gate (AG) block is used to effectively suppress feature activation in uncorrelated regions and highlight the target area. Results: The proposed model is first evaluated on the public dataset DRIVE, on which it performs significantly better than the basic framework in terms of sensitivity (SE), intersection‐over‐union (IOU), and area under the receiver operating characteristic curve (AUC). In particular, the SE and IOU are observed to increase by 7.46% and 5.97%, respectively. Further, the evaluation indices exhibit improvements compared to those of state‐of‐the‐art methods as well, with SE and IOU increasing by 3.58% and 3.26%, respectively. Additionally, in order to demonstrate the generalizability of the proposed architecture, we evaluate our model on three other challenging datasets. The respective performances are observed to be better than those of state‐of‐the‐art network architectures on the same datasets. Moreover, we use lung segmentation as a comparative experiment to demonstrate the transferability of the advantageous properties of the proposed approach in the context of small target segmentation to the segmentation of large targets. Finally, an ablation study is conducted to investigate the individual contributions of the AG block, the DAC block, and the RMP block to the performance of the network. Conclusions: The proposed method is evaluated on various datasets. Experimental results demonstrate that the proposed model performs better than state‐of‐the‐art methods in medical image segmentation of small targets. [ABSTRACT FROM AUTHOR]

In recent years, supervised deep learning methods have shown a great promise in dense depth estimation. However, massive high-quality training data are expensive and impractical to acquire. Alternatively, self-supervised learning-based depth estimators can learn the latent transformation from monocular or binocular video sequences by minimizing the photometric warp error between consecutive frames, but they suffer from the scale ambiguity problem or have difficulty in estimating precise pose changes between frames. In this paper, we propose a joint self-supervised deep learning pipeline for depth and ego-motion estimation by employing the advantages of adversarial learning and joint optimization with spatial-temporal geometrical constraints. The stereo reconstruction error provides the spatial geometric constraint to estimate the absolute scale depth. Meanwhile, the depth map with an absolute scale and a pre-trained pose network serves as a good starting point for direct visual odometry (DVO). DVO optimization based on spatial geometric constraints can result in a fine-grained ego-motion estimation with the additional backpropagation signals provided to the depth estimation network. Finally, the spatial and temporal domain-based reconstructed views are concatenated, and the iterative coupling optimization process is implemented in combination with the adversarial learning for accurate depth and precise ego-motion estimation. The experimental results show superior performance compared with state-of-the-art methods for monocular depth and ego-motion estimation on the KITTI dataset and a great generalization ability of the proposed approach. [ABSTRACT FROM AUTHOR]

Sketch-to-photo recognition is an important challenge in face recognition because it requires matching face images in different domains. Even the deep learning, which has recently been deployed in face recognition, is not efficient for face sketch recognition due to the limited sketch datasets. In this paper, we propose a novel face sketch recognition approach based on transfer learning. We design a three-channel convolutional neural network architecture in which the triplet loss is adopted in order to learn discriminative features and reduce intra-class variations. Moreover, we propose a hard triplet sample selection strategy to augment the number of training samples and avoid slow convergence. With the proposed method, facial features from digital photos and from sketches taken from the same person are closer; the opposite occurs if the digital photo and sketch are from different identities. Experimental results on multiple public datasets indicate that the proposed face sketch recognition method outperforms the existing approaches. [ABSTRACT FROM AUTHOR]

Nowadays, deep learning is widely used to detect interest points and extract the corresponding descriptors and achieved suitable results for many applications of computer vision, such as image matching, three-dimensional reconstruction, simultaneous localization, and mapping. We propose an approach for interest point detection and descriptor extraction using pyramid convolution and circle loss, which is named as PC-SuperPoint. We utilize pyramid convolutions in the backbone network, which includes convolution kernels of different scales for multiscale feature extraction. The following well-designed networks are able to capture the local and global information from the obtained backbone feature maps. In addition, circle loss, which enhances weight attributes for each pair of descriptors, is also applied to improve the convergence speed in the training phase. Experiments on the HPatches dataset and KITTI dataset achieve promising results, which reveal the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Make and model recognition (MMR) of vehicles plays an important role in automatic vision-based systems. This paper proposes a novel deep learning approach for MMR using the SqueezeNet architecture. The frontal views of vehicle images are first extracted and fed into a deep network for training and testing. The SqueezeNet architecture with bypass connections between the Fire modules, a variant of the vanilla SqueezeNet, is employed for this study, which makes our MMR system more efficient. The experimental results on our collected large-scale vehicle datasets indicate that the proposed model achieves 96.3% recognition rate at the rank-1 level with an economical time slice of 108.8 ms. For inference tasks, the deployed deep model requires less than 5 MB of space and thus has a great viability in real-time applications. [ABSTRACT FROM AUTHOR]

Due to efficient photometric information utilization, direct visual odometry (DVO) is getting widely used to estimate the ego-motion of moving cameras as well as map the environment from videos simultaneously, especially in challenging weak-texture scenarios. However, DVO suffers from brightness discrepancies since it directly utilizes intensity patterns of pixels to register frames for camera pose estimation. Most existing brightness transfer methods build a fixed transfer function which is inappropriate for successive and inconsistent brightness changes in practice. To overcome this problem, we propose a Photometric Transfer Net (PTNet) which is trained to pixel-wisely remove brightness discrepancies between two frames without ruining the context information. Photometric consistency in DVO is obtained by adjusting the source frame according to the reference frame. Since no dataset is available for training the photometric transfer model, we augment the EuRoC dataset by generating a certain number of frames with different brightness levels for each original frame by a nonlinear transformation. Afterwards, required training data containing various brightness changes and scene movements along with ground truth can be collected from the extended sequences. Evaluations on both real-world and synthetic datasets demonstrate the effectiveness of the proposed model. Assessment on an unseen dataset with fixed model parameters trained on another dataset proves the generalization ability of the model. Furthermore, we embed the model into DVO to preprocess input data with brightness discrepancies. Experimental results show that PTNet-based DVO achieves more robust initialization and accurate pose estimation than the original one. [ABSTRACT FROM AUTHOR]