An artificial intelligence-based prognostic prediction model for hemorrhagic stroke.

PURPOSE: The prognosis following a hemorrhagic stroke is usually extremely poor. Rating scales have been developed to predict the outcomes of patients with intracerebral hemorrhage (ICH). To date, however, the prognostic prediction models have not included the full range of relevant imaging features. We constructed a clinic-imaging fusion model based on convolutional neural networks (CNN) to predict the short-term prognosis of ICH patients. MATERIALS AND METHODS: This was a multi-center retrospective study, which included 1990 patients with ICH. Two CNN-based deep learning models were constructed to predict the neurofunctional outcomes at discharge; these were validated using a nested 5-fold cross-validation approach. The models' predictive efficiency was compared with the original ICH scale and the ICH grading scale. Poor neurological outcome was defined as a Glasgow Outcome Scale (GOS) score of 1-3. RESULTS: The training and test sets included 1599 and 391 patients, respectively. For the test set, the clinic-imaging fusion model had the highest area under the curve (AUC = 0.903), followed by the imaging-based model (AUC = 0.886), the ICH scale (AUC = 0.777), and finally the ICH grading scale (AUC = 0.747). CONCLUSION: The CNN prognostic prediction model based on neuroimaging features was more effective than the ICH scales in predicting the neurological outcomes of ICH patients at discharge. The CNN model's predictive efficiency slightly improved when clinical data were included.

Phase Interface Regulating on Amorphous/Crystalline Bismuth Catalyst for Boosted Electrocatalytic CO2 Reduction to Formate.

Electroreduction of carbon dioxide into readily collectable and high-value carbon-based fuels is greatly significant to overcome the energy and environmental crises yet challenging in the development of robust and highly efficient electrocatalysts. Herein, a bismuth (Bi) heterophase electrode with enriched amorphous/crystalline interfaces was fabricated via cathodically in situ transformation of Bi-based metal-phenolic complexes (Bi-tannic acid, Bi-TA). Compared with amorphous or crystalline Bi catalyst, the amorphous/crystalline structure Bi leads to significantly enhanced performance for CO2 electroreduction. In a liquid-phase H-type cell, the Faraday efficiency (FE) of formate formation is over 90% in a wide potential range from -0.8 to -1.3 V, demonstrating a high selectivity toward formate. Moreover, in a flow cell, a large current density reaching 600 mA cm-2 can further be rendered for formate production. Theoretical calculations indicate that the amorphous/crystalline Bi heterophase interface exhibits a favorable adsorption of CO2 and lower energy barriers for the rate-determining step compared with the crystalline Bi counterparts, thus accelerating the reaction process. This work paves the way for the rational design of advanced heterointerface catalysts for CO2 reduction.

Standardized measurement of mid-surface shift of brain based on deep Hough transform.

The measurement of mid-surface shift (MSS), the geometric displacement between the actual mid-surface and the ideal midsagittal plane (iMSP), is of great significance for accurate diagnosis, treatment and prognosis of patients with intracranial hemorrhage (ICH). Most previous studies are subject to inherent inaccuracy on account of calculating midline shift (MLS) based on 2D slices and ignoring pathological conditions. In this study, we propose a novel standardized measurement model to quantify the distance and the overall volume of mid-surface shift (MSS-D, MSS-V). Our work has four highlights. First, we develop an end-to-end network architecture with multiple sub-tasks including the actual mid-surface segmentation, hematoma segmentation and iMSP detection, which significantly improves the efficiency and accuracy of MSS measurement by taking advantage of the common properties among tasks. Second, an efficient iMSP detection scheme is proposed based on the differentiable deep Hough transform (DHT), which converts and simplifies the plane detection problem in the image space into a keypoint detection problem in the Hough space. Third, we devise a sparse DHT strategy and a weighted least square (WLS) method to increase the sparsity of features, improving inference speed and greatly reducing computation cost. Fourth, we design a joint loss function to comprehensively consider the correlation of features between multi-tasks and multi-domains. Extensive validation on our large in-house dataset (519 patients) and the public CQ500 dataset (491 patients), demonstrates the superiority of our method over the state-of-the-art methods.

Near-Infrared II Hyperspectral Imaging Improves the Accuracy of Pathological Sampling of Multiple Cancer Types.

Pathological histology is the "gold standard" for clinical diagnosis of cancer. Incomplete or excessive sampling of the formalin-fixed excised cancer specimen will result in inaccurate histologic assessment or excessive workload. Conventionally, pathologists perform specimen sampling relying on naked-eye observation, which is subjective and limited by human perception. Precise identification of cancer tissue, size, and margin is challenging, especially for lesions with inconspicuous tumors. To overcome the limits of human eye perception (visible: 400-700 nm) and improve the sampling efficiency, in this study, we propose using a second near-infrared window (NIR-II: 900-1700 nm) hyperspectral imaging (HSI) system to assist specimen sampling on the strength of the verified deep anatomical penetration and low scattering characteristics of the NIR-II optical window. We used selected NIR-II HSI narrow bands to synthesize color images for human eye observation and also applied a machine learning-based algorithm on the complete NIR-II HSI data for automatic tissue classification to assist pathologists in specimen sampling. A total of 92 tumor samples were collected, including 7 types. Sixty-two (62/92) samples were used as the validation set. Five experienced pathologists marked the contour of the cancer tissue on conventional color images by using different methods, and compared it with the "gold standard," showing that NIR-II HSI-assisted methods had significant improvements in determining cancer tissue compared with conventional methods (conventional color image with or without X-ray). The proposed system can be easily integrated into the current workflow, with high imaging efficiency and no ionizing radiation. It may also find applications in intraoperative detection of residual lesions and identification of different tissues.

SC-AIR-BERT: a pre-trained single-cell model for predicting the antigen-binding specificity of the adaptive immune receptor.

Accurately predicting the antigen-binding specificity of adaptive immune receptors (AIRs), such as T-cell receptors (TCRs) and B-cell receptors (BCRs), is essential for discovering new immune therapies. However, the diversity of AIR chain sequences limits the accuracy of current prediction methods. This study introduces SC-AIR-BERT, a pre-trained model that learns comprehensive sequence representations of paired AIR chains to improve binding specificity prediction. SC-AIR-BERT first learns the 'language' of AIR sequences through self-supervised pre-training on a large cohort of paired AIR chains from multiple single-cell resources. The model is then fine-tuned with a multilayer perceptron head for binding specificity prediction, employing the K-mer strategy to enhance sequence representation learning. Extensive experiments demonstrate the superior AUC performance of SC-AIR-BERT compared with current methods for TCR- and BCR-binding specificity prediction.

High Dynamic Range Dual-Modal White Light Imaging Improves the Accuracy of Tumor Bed Sampling After Neoadjuvant Therapy for Breast Cancer.

OBJECTIVES: Accurate evaluation of residual cancer burden remains challenging because of the lack of appropriate techniques for tumor bed sampling. This study evaluated the application of a white light imaging system to help pathologists differentiate the components and location of tumor bed in specimens. METHODS: The high dynamic range dual-mode white light imaging (HDR-DWI) system was developed to capture antiglare reflection and multiexposure HDR transmission images. It was tested in 60 specimens of modified radical mastectomy after neoadjuvant therapy. We observed the differential transmittance among tumor tissue, fibrosis tissue, and adipose tissue. RESULTS: The sensitivity and specificity of HDR-DWI were compared with x-ray or visual examination to determine whether HDR-DWI was superior in identifying tumor beds. We found that tumor tissue had lower transmittance (0.12 ± 0.03) than fibers (0.15 ± 0.04) and fats (0.27 ± 0.07) (P < .01). CONCLUSIONS: HDR-DWI was more sensitive in identifying fiber and tumor tissues than cabinet x-ray and visual observation (P < .01). In addition, HDR-DWI could identify more fibrosis areas than the currently used whole slide imaging did in 12 samples (12/60). We have determined that HDR-DWI can provide more in-depth tumor bed information than x-ray and visual examination do, which will help prevent diagnostic errors in tumor bed sampling.

Lymph Node Fibroblastic Reticular Cells Attenuate Immune Responses Through Induction of Tolerogenic Macrophages at Early Stage of Transplantation.

BACKGROUND: Fibroblastic reticular cells (FRCs) are a type of stromal cells located in the T zone in secondary lymphoid organs. Previous studies showed that FRCs possess the potential to promote myeloid differentiation. We aim to investigate whether FRCs in lymph nodes (LNs) could induce tolerogenic macrophage generation and further influence T-cell immunity at an early stage of allogeneic hematopoietic stem cell transplantation (allo-HSCT). METHODS: LNs were assayed to confirm the existence of proliferating macrophages after allo-HSCT. Ex vivo-expanded FRCs and bone marrow cells were cocultured to verify the generation of macrophages. Real-time quantitative PCR and ELISA assays were performed to observe the cytokines expressed by FRC. Transcriptome sequencing was performed to compare the difference between FRC-induced macrophages (FMs) and conventional macrophages. Mixed lymphocyte reaction and the utilization of FMs in acute graft-versus-host disease (aGVHD) mice were used to test the inhibitory function of FMs in T-cell immunity in vitro and in vivo. RESULTS: We found a large number of proliferating macrophages near FRCs in LNs with tolerogenic phenotype under allo-HSCT conditions. Neutralizing anti-macrophage colony-stimulating factor receptor antibody abolished FMs generation in vitro. Phenotypic analysis and transcriptome sequencing suggested FMs possessed immunoinhibitory function. Mixed lymphocyte reaction proved that FMs could inhibit T-cell activation and differentiation toward Th1/Tc1 cells. Injection of FMs in aGVHD mice effectively attenuated aGVHD severity and mortality. CONCLUSIONS: This study has revealed a novel mechanism of immune regulation through the generation of FRC-induced tolerogenic macrophages in LNs at an early stage of allo-HSCT.

A machine learning approach for predicting perihematomal edema expansion in patients with intracerebral hemorrhage.

OBJECTIVES: Preventing the expansion of perihematomal edema (PHE) represents a novel strategy for the improvement of neurological outcomes in intracerebral hemorrhage (ICH) patients. Our goal was to predict early and delayed PHE expansion using a machine learning approach. METHODS: We enrolled 550 patients with spontaneous ICH to study early PHE expansion, and 389 patients to study delayed expansion. Two imaging researchers rated the shape and density of hematoma in non-contrast computed tomography (NCCT). We trained a radiological machine learning (ML) model, a radiomics ML model, and a combined ML model, using data from radiomics, traditional imaging, and clinical indicators. We then validated these models on an independent dataset by using a nested 4-fold cross-validation approach. We compared models with respect to their predictive performance, which was assessed using the receiver operating characteristic curve. RESULTS: For both early and delayed PHE expansion, the combined ML model was most predictive (early/delayed AUC values were 0.840/0.705), followed by the radiomics ML model (0.799/0.663), the radiological ML model (0.779/0.631), and the imaging readers (reader 1: 0.668/0.565, reader 2: 0.700/0.617). CONCLUSION: We validated a machine learning approach with high interpretability for the prediction of early and delayed PHE expansion. This new technique may assist clinical practice for the management of neurocritical patients with ICH. KEY POINTS: • This is the first study to use artificial intelligence technology for the prediction of perihematomal edema expansion. • A combined machine learning model, trained on data from radiomics, clinical indicators, and imaging features associated with hematoma expansion, outperformed all other methods.

Hierarchically hollow N-doped carbon-cobalt nanoparticle heterointerface for efficient bifunctional oxygen electrocatalysis.

The rational design and fabrication of high-performance and durable bifunctional non-noble-metal electrocatalysts for both the oxygen reduction reaction (ORR) and oxygen evolution reaction (OER) are still a great challenge in the practical applications of rechargeable zinc-air (Zn-air) batteries. Herein, we report a simple yet robust route to synthesize cobalt nanoparticles rooted in the hierarchically hollow nitrogen-doped carbon frameworks (Co@HNCs). This strategy employs the pyrolysis of nanostructured hollow Co-based metal-organic framework (ZIF-67) precursors produced by selective linker cleaving with pyrazino(2,3-f)(1,10)phenanthroline-2,3-dicarboxylic acid molecules (H2PPDA). The designed hierarchically architecture is favorable for the accessibility of the active sites in the catalyst, which affords enhanced bifunctional performance for ORR and OER. Moreover, when used as a cathode in liquid and all-solid-state Zn-air batteries, the resultant Co@HNCs delivers high efficiency and outstanding durability, even outperforming the benchmark Pt/C + RuO2. This work provides a feasible design avenue to achieve advanced dual-phasic oxygen electrocatalyst and promotes the development of rechargeable Zn-air batteries.

Defining Delayed Perihematomal Edema Expansion in Intracerebral Hemorrhage: Segmentation, Time Course, Risk Factors and Clinical Outcome.

We attempt to generate a definition of delayed perihematomal edema expansion (DPE) and analyze its time course, risk factors, and clinical outcomes. A multi-cohort data was derived from the Chinese Intracranial Hemorrhage Image Database (CICHID). A non-contrast computed tomography (NCCT) -based deep learning model was constructed for fully automated segmentation hematoma and perihematomal edema (PHE). Time course of hematoma and PHE evolution correlated to initial hematoma volume was volumetrically assessed. Predictive values for DPE were calculated through receiver operating characteristic curve analysis and were tested in an independent cohort. Logistic regression analysis was utilized to identify risk factors for DPE formation and poor outcomes. The test cohort's Dice scores of lesion segmentation were 0.877 and 0.642 for hematoma and PHE, respectively. Overall, 1201 patients were enrolled for time-course analysis of ICH evolution. A total of 312 patients were further selected for DPE analysis. Time course analysis showed the growth peak of PHE approximately concentrates in 14 days after onset. The best cutoff for DPE to predict poor outcome was 3.34 mL of absolute PHE expansion from 4-7 days to 8-14 days (AUC=0.784, sensitivity=72.2%, specificity=81.2%), and 3.78 mL of absolute PHE expansion from 8-14 days to 15-21 days (AUC=0.682, sensitivity=59.3%, specificity=92.1%) in the derivation sample. Patients with DPE was associated with worse outcome (OR: 12.340, 95%CI: 6.378-23.873, P<0.01), and the larger initial hematoma volume (OR: 1.021, 95%CI: 1.000-1.043, P=0.049) was the significant risk factor for DPE formation. This study constructed a well-performance deep learning model for automatic segmentations of hematoma and PHE. A new definition of DPE was generated and is confirmed to be related to poor outcomes in ICH. Patients with larger initial hematoma volume have a higher risk of developing DPE formation.

Reconstructing virtual large slides can improve the accuracy and consistency of tumor bed evaluation for breast cancer after neoadjuvant therapy.

BACKGROUND: To explore whether the "WSI Stitcher", a program we developed for reconstructing virtual large slide through whole slide imaging fragments stitching, can improve the efficiency and consistency of pathologists in evaluating the tumor bed after neoadjuvant treatment of breast cancer compared with the conventional methods through stack splicing of physical slides. METHODS: This study analyzed the advantages of using software-assisted methods to evaluate the tumor bed after neoadjuvant treatment of breast cancer. This new method is to use "WSI Stitcher" to stitch all the WSI fragments together to reconstruct a virtual large slide and evaluate the tumor bed with the help of the built-in ruler and tumor proportion calculation functions. RESULTS: Compared with the conventional method, the evaluation time of the software-assisted method was shortened by 35%(P < 0.001). In the process of tumor bed assessment after neoadjuvant treatment of breast cancer, the software-assisted method has higher intraclass correlation coefficient when measuring the length (0.994 versus 0.934), width (0.992 versus 0.927), percentage of residual tumor cells (0.947 versus 0.878), percentage of carcinoma in situ (0.983 versus 0.881) and RCB index(0.997 versus 0.772). The software-assisted method has higher kappa values when evaluating tumor staging(0.901 versus 0.687) and RCB grading (0.963 versus 0.857). CONCLUSION: The "WSI Stitcher" is an effective tool to help pathologists with the assessment of breast cancer after neoadjuvant treatment.

Automatic Brain Midline Surface Delineation on 3D CT Images With Intracranial Hemorrhage.

Brain midline delineation plays an important role in guiding intracranial hemorrhage surgery, which still remains a challenging task since hemorrhage shifts the normal brain configuration. Most previous studies detected brain midline on 2D plane and did not handle hemorrhage cases well. We propose a novel and efficient hemisphere-segmentation framework (HSF) for 3D brain midline surface delineation. Specifically, we formulate the brain midline delineation as a 3D hemisphere segmentation task, and employ an edge detector and a smooth regularization loss to generate the midline surface. We also introduce a distance-weighted map to keep the attention on the midline. Furthermore, we adopt rectification learning to handle various head poses. Finally, considering the complex situation of ventricle break-in for hemorrhages in bilateral intraventricular (B-IVH) cases, we identify those cases via a classification model and design a midline correction strategy to locally adjust the midline. To our best knowledge, it is the first study focusing on delineating the brain midline surface on 3D CT images of hemorrhage patients and handling the situation of ventricle break-in. Extensive validation on our large in-house datasets (519 patients) and the public CQ500 dataset (491 patients), demonstrates that our method outperforms state-of-the-art methods on brain midline delineation.

Corrigendum: LYG1 Deficiency Attenuates the Severity of Acute Graft-Versus-Host Disease via Skewing Allogeneic T Cells Polarization Towards Treg Cells.

[This corrects the article DOI: 10.3389/fimmu.2021.647894.].

LYG1 Deficiency Attenuates the Severity of Acute Graft-Versus-Host Disease via Skewing Allogeneic T Cells Polarization Towards Treg Cells.

Acute graft-versus-host disease (aGVHD) is a lethal complication after allogeneic hematopoietic stem cell transplantation. The mechanism involves the recognition of host antigens by donor-derived T cells which induces augmented response of alloreactive T cells. In this study, we characterized the role of a previously identified novel classical secretory protein with antitumor function-LYG1 (Lysozyme G-like 1), in aGVHD. LYG1 deficiency reduced the activation of CD4+ T cells and Th1 ratio, but increased Treg ratio in vitro by MLR assay. By using major MHC mismatched aGVHD model, LYG1 deficiency in donor T cells or CD4+ T cells attenuated aGVHD severity, inhibited CD4+ T cells activation and IFN-γ expression, promoted FoxP3 expression, suppressed CXCL9 and CXCL10 expression, restrained allogeneic CD4+ T cells infiltrating in target organs. The function of LYG1 in aGVHD was also confirmed using haploidentical transplant model. Furthermore, administration of recombinant human LYG1 protein intraperitoneally aggravated aGVHD by promoting IFN-γ production and inhibiting FoxP3 expression. The effect of rhLYG1 could partially be abrogated with the absence of IFN-γ. Furthermore, LYG1 deficiency in donor T cells preserved graft-versus-tumor response. In summary, our results indicate LYG1 regulates aGVHD by the alloreactivity of CD4+ T cells and the balance of Th1 and Treg differentiation of allogeneic CD4+ T cells, targeting LYG1 maybe a novel therapeutic strategy for preventing aGVHD.

The Potential Genes Mediate the Pathogenicity of Allogeneic CD4+T Cell in aGVHD Mouse Model.

Acute graft-versus-host disease (aGVHD) remains a significant and severe complication of allogeneic hematopoietic stem cell transplantation (allo-HSCT). Due to the occurrence of aGVHD, allo-HSCT significantly increases the mortality rate compared with autologous hematopoietic stem cell transplantation (auto-HSCT). In this study, auto-HSCT and allo-HSCT aGVHD mouse models were built to detect the difference in CD4+ lymphocyte in different tissues based on ribonucleic acid sequencing (RNA-Seq) analysis. Clustering analysis, functional annotation, and pathway enrichment analysis were performed on differentially expressed genes (DEGs). The protein-protein interaction (PPI) network was used to find hub genes. CD4+T cells were activated by MLR and cytokine stimulation. Cells were sorted out by a flow cell sorter. The selected genes were verified by qRT-PCR, histology, and immunofluorescence staining. The GSE126518 GEO dataset was used to verify the hub genes. Enrichment analysis revealed four immune-related pathways that play an important role in aGVHD, including immunoregulatory interactions between a lymphoid and a nonlymphoid cell, chemokine receptors binding chemokines, cytokine and cytokine receptor interaction, and the chemokine signaling pathway. At the same time, with the PPI network, 11 novel hub genes that were most likely to participate in immunoregulation in aGVHD were identified, which were further validated by qRT-PCR and the GSE126518 dataset. Besides, the protein expression level of Cxcl7 was consistent with the sequencing results. In summary, this study revealed that immunoregulation-related DEGs and pathways played a vital role in the onset of aGVHD. These findings may provide some new clues for probing the pathogenesis and treatment of aGVHD.

Joint affine and deformable three-dimensional networks for brain MRI registration.

PURPOSE: Volumetric medical image registration has important clinical significance. Traditional registration methods may be time-consuming when processing large volumetric data due to their iterative optimizations. In contrast, existing deep learning-based networks can obtain the registration quickly. However, most of them require independent rigid alignment before deformable registration; these two steps are often performed separately and cannot be end-to-end. METHODS: We propose an end-to-end joint affine and deformable network for three-dimensional (3D) medical image registration. The proposed network combines two deformation methods; the first one is for obtaining affine alignment and the second one is a deformable subnetwork for achieving the nonrigid registration. The parameters of the two subnetworks are shared. The global and local similarity measures are used as loss functions for the two subnetworks, respectively. Moreover, an anatomical similarity loss is devised to weakly supervise the training of the whole registration network. Finally, the trained network can perform deformable registration in one forward pass. RESULTS: The efficacy of our network was extensively evaluated on three public brain MRI datasets including Mindboggle101, LPBA40, and IXI. Experimental results demonstrate our network consistently outperformed several state-of-the-art methods with respect to the metrics of Dice index (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD). CONCLUSIONS: The proposed network provides accurate and robust volumetric registration without any pre-alignment requirement, which facilitates the end-to-end deformable registration.

Unsupervised 3D End-to-end Deformable Network for Brain MRI Registration.

Volumetric medical image registration has important clinical significance. Traditional registration methods may be time-consuming when processing large volumetric data due to their iterative optimizations. In contrast, existing deep learning-based networks can obtain the registration quickly. However, most of them require independent rigid alignment before deformable registration; these two steps are often performed separately and cannot be end-to-end. Moreover, registration ground-truth is difficult to obtain for supervised learning methods. To tackle the above issues, we propose an unsupervised 3D end-to-end deformable registration network. The proposed network cascades two subnetworks; the first one is for obtaining affine alignment, and the second one is a deformable subnetwork for achieving the non-rigid registration. The parameters of the two subnetworks are shared. The global and local similarity measures are used as loss functions for the two subnetworks, respectively. The trained network can perform end-to-end deformable registration. We conducted experiments on brain MRI datasets (LPBA40, Mindboggle101, and IXI) and experimental results demonstrate the efficacy of the proposed registration network.

In Vitro Immunological Effects of CXCR3 Inhibitor AMG487 on Dendritic Cells.

AMG 487 is the targeted blocker of chemokine receptor CXCR3 and improves inflammatory symptoms by blocking the inflammatory cycle. Here we investigated whether AMG 487 affects dendritic cell (DC) biology and function. The expression of co-stimulatory markers on DCs was reduced, indicating the semi-mature state of DC when AMG 487 was added throughout the in vitro differentiation period. Additionally, when added solely during the final lipopolysaccharide-induced activation step, AMG 487 inhibited DC activation, as demonstrated by a decreased expression of activation markers. AMG487 also promoted the expression of PD-L2 and impaired the ability to induce antigen-specific T cell responses. Our results demonstrated that AMG 487 significantly affects DC maturity in vitro and function leading to impaired T cell activation, inducing DCs to have characteristics similar to tolerogenic DCs. AMG 487 may directly play an immunomodulatory role during DC development and functional shaping.

Deeply-Supervised Networks With Threshold Loss for Cancer Detection in Automated Breast Ultrasound.

ABUS, or Automated breast ultrasound, is an innovative and promising method of screening for breast examination. Comparing to common B-mode 2D ultrasound, ABUS attains operator-independent image acquisition and also provides 3D views of the whole breast. Nonetheless, reviewing ABUS images is particularly time-intensive and errors by oversight might occur. For this study, we offer an innovative 3D convolutional network, which is used for ABUS for automated cancer detection, in order to accelerate reviewing and meanwhile to obtain high detection sensitivity with low false positives (FPs). Specifically, we offer a densely deep supervision method in order to augment the detection sensitivity greatly by effectively using multi-layer features. Furthermore, we suggest a threshold loss in order to present voxel-level adaptive threshold for discerning cancer vs. non-cancer, which can attain high sensitivity with low false positives. The efficacy of our network is verified from a collected dataset of 219 patients with 614 ABUS volumes, including 745 cancer regions, and 144 healthy women with a total of 900 volumes, without abnormal findings. Extensive experiments demonstrate our method attains a sensitivity of 95% with 0.84 FP per volume. The proposed network provides an effective cancer detection scheme for breast examination using ABUS by sustaining high sensitivity with low false positives. The code is publicly available at https://github.com/nawang0226/abus_code.

Edge-Aware Pyramidal Deformable Network for Unsupervised Registration of Brain MR Images.

Deformable image registration is of essential important for clinical diagnosis, treatment planning, and surgical navigation. However, most existing registration solutions require separate rigid alignment before deformable registration, and may not well handle the large deformation circumstances. We propose a novel edge-aware pyramidal deformable network (referred as EPReg) for unsupervised volumetric registration. Specifically, we propose to fully exploit the useful complementary information from the multi-level feature pyramids to predict multi-scale displacement fields. Such coarse-to-fine estimation facilitates the progressive refinement of the predicted registration field, which enables our network to handle large deformations between volumetric data. In addition, we integrate edge information with the original images as dual-inputs, which enhances the texture structures of image content, to impel the proposed network pay extra attention to the edge-aware information for structure alignment. The efficacy of our EPReg was extensively evaluated on three public brain MRI datasets including Mindboggle101, LPBA40, and IXI30. Experiments demonstrate our EPReg consistently outperformed several cutting-edge methods with respect to the metrics of Dice index (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD). The proposed EPReg is a general solution for the problem of deformable volumetric registration.