Fast and accurate segmentation of stroke lesions is highly desirable to help specialists in lesion measurements and making treatment plans. Segmentation of stroke lesions are challenging due to evolvement of stroke, low contrast and highly variable grayscale distributions. In this paper, we propose a novel multi-task learning framework to achieve enhanced segmentation of stroke lesions. In this framework, without increasing the inference time and model parameters in the inference phase, we perform supervised learning on a main task (stroke lesion segmentation) and an auxiliary task (edge segmentation of the brain lesion) to improve the performance of the main task. For validation, a CT brain infarction dataset has been constructed. In terms of the Dice coefficient, the proposed method could enhance at least 4.76% on the constructed CT dataset, and at least 7.97% on the open source dataset Anatomical Tracings of Lesions After Stroke than state-of-the-art models.

With the development of deep learning, deep convolution neural networks for medical image segmentation tasks have become more and more complex in pursuit of higher accuracy. In most scenarios, medical image segmentation pursues accuracy rather than speed, However, real-time performance is crucial in some scenarios, such as surgical navigation and diagnosis of acute stroke. So design of high-precision, lightweight and real-time medical image segmentation network has become an urgent need. To this end, a novel lightweight dual-domain network (LDD-Net) has been proposed in this paper. LDD-Net is comprised of two branches, learning respectively from the frequency domain and the spatial domain. In the frequency domain branch, the image spatial resolution is compressed via discrete cosine transform to have a large receptive field, so that better semantic context features can be learned. In the spatial domain branch, high-resolution feature representations with more details are learned. Finally, the learned features of these two branches are fused to yield high accuracy with low computational cost. The proposed method has been validated on two medical image segmentation datasets to yield the state-of-the-art performances with greatly reduced inference time and parameters of the learned models.

Previous generic object detection networks depend on complex structures and huge computational costs, which are not conducive to deployment and application in resource-constrained scenarios, especially in mobile devices. Accordingly, exploring lightweight network has become a focus issue, attracting the widespread attention of industry and academia. In this paper, we proposed a lightweight two-stage object detection framework based convolutional neural network, which includes two components: the feature extractor and the detection components. For the feature extractor component, two architecture modules were proposed: residual cross-branch feature extraction module (RCM) and multi-scale feature fusion module (MFM). The RCM module adopted residual structure and the mechanism of cross-branch multi-channel to increase the receptive field and expression capability for low-level features. The MFM module integrated feature maps with multiple scales to enhance global context information. Moreover, for further optimizing the distribution of features and paying more attention to the detection-related regions, a separate channel self-attention module was proposed. Extensive experiments on the public datasets PASCAL VOC illustrated the superiority of the proposed model.

X-ray baggage security image inspection is crucial for maintaining safety and is difficult due to the 2D projection nature of the images and variable sizes of objects. We proposed an anchor-free symmetric triangle network (ST-Net) for object detection within X-ray baggage security imagery. The ST-Net has three main components: bottom-up structure, Symmetric triangle feature pyramid module (STFPM) and detector head. Specifically, the STFPM module has multiple paths with different directions and lengths, which can fuse the rich multi-scale feature representation. Symmetry structure can effectively supplement the global feature information. In addition, we built a new dual-energy (DE) dataset generated from the real dual-energy images to preserve the original information. Experiments on DE dataset and public SIXray dataset demonstrate the proposed ST-Net could surpass most existing methods.

Although electroconvulsive therapy (ECT) is one of the most effective treatments for major depressive disorder (MDD), the mechanism underlying the therapeutic efficacy and side effects of ECT remains poorly understood. Here, we investigated alterations in the cortical morphological measurements including cortical thickness (CT), surface area (SA), and local gyrification index (LGI) in 23 MDD patients before and after ECT. Furthermore, multivariate pattern analysis using linear support vector machine (SVM) was applied to investigate whether the changed morphological measurements can be effective indicators for therapeutic efficacy of ECT. Surface-based morphometry (SBM) analysis found significantly increased vertex-wise and regional cortical thickness (CT) and surface area (SA) in widespread regions, mainly located in the left insula (INS) and left fusiform gyrus, as well as hypergyrification in the left middle temporal gyrus (MTG) in MDD patients after ECT. Partial correlational analyses identified associations between the morphological properties and depressive symptom scores and impaired memory scores. Moreover, SVM result showed that the changed morphological measurements were effective to classify the MDD patients before and after ECT. Our findings suggested that ECT may enhance cortical neuroplasticity to facilitate neurogenesis to remit depressive symptoms and to impair delayed memory. These findings indicated that the cortical morphometry is a good index for therapeutic efficacy of ECT.

A fast and robust algorithm to segment the head from T1-weighted magnetic resonance (MR) volumes is presented based on robust determination of threshold from combination of the original volume (through employing the fuzzy c-means clustering which is insensitive to intensity inhomogeneity) and the average axial image (which suppresses the intensity noise), efficient connected component analysis through iterative recursion and fast implementation of morphological operations via decomposition of structuring elements. The algorithm has been validated against 101 data sets with various artifacts

This paper presents a method to improve the semi-automatic method for intensity inhomogeneity correction by Dawant et al. through introducing a fully automatic approach to reference points generation, which is based on order statistics and integrates information from the fine to coarse scale representations of the input image. The method has been validated and compared with two popular methods, N3 and BFC. Advantages of the proposed method are demonstrated.

A novel thresholding approach to confine the intensity frequency range of the object based on supervision is introduced. It consists of three steps. First, the region of interest (ROI) is determined in the image. Then, from the histogram of the ROI, the frequency range in which the proportion of the background to the ROI varies is estimated through supervision. Finally, the threshold is determined by minimizing the classification error within the constrained variable background range. The performance of the approach has been validated against 54 brain MR images, including images with severe intensity inhomogeneity and/or noise, CT chest images, and the Cameraman image. Compared with nonsupervised thresholding methods, the proposed approach is substantially more robust and more reliable. It is also computationally efficient and can be applied to a wide class of computer vision problems, such as character recognition, fingerprint identification, and segmentation of a wide variety of medical images.

An algorithm to locate the four modified Talairach cortical landmarks (A, P, L, and R) based on knowledge is proposed. Knowledge is employed to select the axial plane (the AP plane) passing through the anterior and posterior commissures, determination of optimum thresholds, segmentation, and refinement of the AP plane. The algorithm has been validated against 38 T1-weighted datasets from public resources. It takes less than 1 second on Pentium 4 (2.6 GHz) to extract the 4 landmarks. The average landmark location error is below 1 mm. The algorithm is robust and accurate as the factors influencing the determination of cortical landmarks like noise, gray level inhomogeneity, partial volume effect, and connection between the cortex and the sagittal sinus/meninges/skull are carefully compensated. A low computational cost is attributed to simple operations like thresholding, seeding, simple morphological operations and distance transform.

In this paper, we propose an automatic method to segment the Sylvian fissure (SF) from T1-weighted magnetic resonance (MR) images. The method consists of four steps. First, morphological methods are applied to remove non-brain tissues. Then, 2D region growing is used to segment the cerebral spinal fluid (CSF) and gray matter (GM) in the SF, in which spatial relationships are used to control the region growing. A modified fuzzy c-means (FCM) clustering algorithm is introduced as well, which is more accurate than the conventional FCM method. After that, we get the 3D volume of the SF through segmenting CSF and GM slice by slice. Finally, CSF in the SF is extracted incorporating the anatomical information on a constant cortex thickness. We have demonstrated our method through both phantom and patient data.

This paper presents a knowledge-driven algorithm to identify and segment the central sulcus (CS) from human brain MR images. The dataset is reformatted along the anterior and posterior commissures (AC-PC) plane first. Then, the 3D region within the two coronal planes passing through the AC and PC is defined as the region of interest (ROl) to search for all the sulci within it. The CS is the sulcus with the largest volume within the ROI. Together with the sulci, grey matter (GM) is included for the region growing in order to deal with the partial volume effect. The GM is removed through skeletonization. Experimental results are given.