3D Brain Images Research Articles

3D brain image based tumor classification using ensemble of reinforcement transfer-based belief neural networks

3D brain image based tumor classification using ensemble of reinforcement transfer-based belief neural networks

Spatio-temporal brain invasion pattern of Streptococcus pneumoniae and dynamic changes in the cellular environment in bacteremia-derived meningitis

Streptococcus pneumoniae (the pneumococcus) is the major cause of bacterial meningitis globally, and pneumococcal meningitis is associated with increased risk of long-term neurological sequelae. These include several sensorimotor functions that are controlled by specific brain regions which, during bacterial meningitis, are damaged by a neuroinflammatory response and the deleterious action of bacterial toxins in the brain. However, little is known about the invasion pattern of the pneumococcus into the brain. Using a bacteremia-derived meningitis mouse model, we combined 3D whole brain imaging with brain microdissection to show that all brain regions were equally affected during disease progression, with the presence of pneumococci closely associated to the microvasculature. In the hippocampus, the invasion provoked microglial activation, while the neurogenic niche showed increased proliferation and migration of neuroblasts. Our results indicate that, even before the outbreak of symptoms, the bacterial load throughout the brain is high and causes neuroinflammation and cell death, a pathological scenario which ultimately leads to a failing regeneration of new neurons.

An MR-based brain template and atlas for optical projection tomography and light sheet fluorescence microscopy in neuroscience.

Optical Projection Tomography (OPT) and light sheet fluorescence microscopy (LSFM) are high resolution optical imaging techniques, ideally suited for ex vivo 3D whole mouse brain imaging. Although they exhibit high specificity for their targets, the anatomical detail provided by tissue autofluorescence remains limited. T1-weighted images were acquired from 19 BABB or DBE cleared brains to create an MR template using serial longitudinal registration. Afterwards, fluorescent OPT and LSFM images were coregistered/normalized to the MR template to create fusion images. Volumetric calculations revealed a significant difference between BABB and DBE cleared brains, leading to develop two optimized templates, with associated tissue priors and brain atlas, for BABB (OCUM) and DBE (iOCUM). By creating fusion images, we identified virus infected brain regions, mapped dopamine transporter and translocator protein expression, and traced innervation from the eye along the optic tract to the thalamus and superior colliculus using cholera toxin B. Fusion images allowed for precise anatomical identification of fluorescent signal in the detailed anatomical context provided by MR. The possibility to anatomically map fluorescent signals on magnetic resonance (MR) images, widely used in clinical and preclinical neuroscience, would greatly benefit applications of optical imaging of mouse brain. These specific MR templates for cleared brains enable a broad range of neuroscientific applications integrating 3D optical brain imaging.

Improving the Quality of CT Images of Stroke Patients Using the SSDE and K-Means Segmentation in the Radiology Installation of RSUD A.W. Sjahranie Samarinda

Stroke examination using a CT scan is a technique that produces 3D brain images without having to do surgery. The purpose of this study was to determine the dose received by each patient who underwent a CT-scan procedure and to improve the image quality of the results. Optimization of the dose for each patient is based on a comparison between the calculated SSDE value and the DRL in the form of CTDIvol and DLP, which are implemented in Indonesia. Improving the quality of CT-Scan images through digital processing using the segmentation method to determine the stroke area. The method used is K-Means segmentation. The results obtained for the SSDE value are the dose received by the patient which is higher than the dose issued by the device or the value. However, the SSDE, CTDIvol, and DLP values are still below the DRL standard range set by BAPETEN, so they are still within the safe inspection threshold. The average CTDIvol value, which is the output dose value of the device from 22 patients, was 41.84 mGy and the 75th percentile value was 42.7 mGy. The average DLP value is 1054.08 mGy.cm and the 75th percentile value of the DLP value is 964.13 mGy.cm. The average value of the SSDE value which is the value of the dose received by the patient is 43.39 mGy and the 75th percentile value is 42.440 mGy. Based on these results, the value of the dose received by the patient is higher than the value of the dose issued by the device. Segmentation using the K-Means method with the number of clusters k = 3 produces an accuracy of 86.3%. This means that the patient receives a dose of CT-Scan radiation within safe limits and his image can be processed properly, as evidenced by the high success of the segmentation process.

Analysis of the effect of motion on highly accelerated 3D FatNavs in 3D brain images acquired at 3T.

3D FatNavs are rapid acquisitions of MRI fat-volumes within the head that can be used for retrospective motion correction for brain MRI. 3D FatNavs typically use very high acceleration factors and are reconstructed with the GRAPPA parallel imaging technique. However, the GRAPPA reconstruction is not expected to perform well on 3D FatNavs volumes in the presence of strong motion due to the mismatched calibration data acquired once at the start of the scan, leading to motion-parameter misestimation. This study aims to assess the accuracy and precision of 3D FatNav-derived motion-estimates in the presence of large changes in head position. Rigid motion parameters were simulated and applied retrospectively to the 3D FatNav volumes from MPRAGE datasets acquired at 3T. The transformed images were then re-reconstructed using GRAPPA to simulate real motion deterioration of the fat-navigator, and used to estimate the motion applied and evaluate the tracking inaccuracy. This information was then used to estimate the residual motion after 3D FatNav-based motion correction and applied to the original MPRAGE volumes. The effect of the misestimation was assessed using an image quality metric and the evaluation scores from two observers. Quality boundaries were then estimated to assess the motion tolerance when 3D FatNavs are used. The GRAPPA reconstruction was shown to deteriorate for large changes in the head position, affecting the quality of 3D FatNav volumes and consequently degrading the accuracy of the motion-estimates. Based on our simulations, the estimated threshold of motion that led to a noticeable degradation in the MPRAGE image quality was up to RMS values of 3.7° and 3 mm for rotations and translations respectively. 3D FatNavs were shown to be able to correct for a wide range of motion levels and types. Boundaries of acceptable motion magnitudes for different levels of acceptable loss of image quality were determined.

Microplastics reach the brain and interfere with honey bee cognition

Scientific research on the impact of microplastics (MPs) in terrestrial systems is still emerging, but it has confirmed adverse health effects in organisms exposed to plastics. Although recent studies have shown the toxicological effects of individual MPs polymers on honey bees, the effects of different polymer combinations on cognitive and behavioural performance remain unknown. To fill this knowledge gap, we investigated the effects of oral exposure to spherical MPs on cognitive performance and brain accumulation in the honey bee Apis mellifera. We evaluated the acute toxicity, after a two-day exposure, of polystyrene (PS - 4.8-5.8 μm) and plexiglass (Poly(methyl methacrylate), or PMMA - 1-40 μm) MPs, and a combination of the two (MIX), at two environmentally relevant and one higher concentration (0.5, 5 and 50 mg L−1) and analysed their effects on sucrose responsiveness and appetitive olfactory learning and memory. We also used fluorescent thermoset amino formaldehyde MPs (1–5 μm) to explore whether microspheres of this diameter could penetrate the insect blood-brain barrier (BBB), using Two-Photon Fluorescence Microscopy (TPFM) in combination with an optimized version of the DISCO clearing technique. The results showed that PS reduced sucrose responsiveness, while PMMA had no significant effect; however, the combination had a marked negative effect on sucrose responsiveness. PMMA, PS, and MIX impaired bee learning and memory in bees, with PS showing the most severe effects. 3D brain imaging analysis using TFPM showed that 1–5 μm MPs penetrated and accumulated in the brain after only three days of oral exposure. These results raise concerns about the potential mechanical, cellular, and biochemical damage that MPs may cause to the central nervous system.

In vivo three-dimensional brain imaging with chemiluminescence probes in Alzheimer's disease models.

Optical three-dimensional (3D) molecular imaging is highly desirable for providing precise distribution of the target-of-interest in disease models. However, such 3D imaging is still far from wide applications in biomedical research; 3D brain optical molecular imaging, in particular, has rarely been reported. In this report, we designed chemiluminescence probes with high quantum yields, relatively long emission wavelengths, and high signal-to-noise ratios to fulfill the requirements for 3D brain imaging in vivo. With assistance from density-function theory (DFT) computation, we designed ADLumin-Xs by locking up the rotation of the double bond via fusing the furan ring to the phenyl ring. Our results showed that ADLumin-5 had a high quantum yield of chemiluminescence and could bind to amyloid beta (Aβ). Remarkably, ADLumin-5's radiance intensity in brain areas could reach 4 × 107 photon/s/cm2/sr, which is probably 100-fold higher than most chemiluminescence probes for in vivo imaging. Because of its strong emission, we demonstrated that ADLumin-5 could be used for in vivo 3D brain imaging in transgenic mouse models of Alzheimer's disease.

Multidimensional compressed sensing to advance 23 Na multi-quantum coherences MRI.

Sodium (23 Na) multi-quantum coherences (MQC) MRI was accelerated using three-dimensional (3D) and a dedicated five-dimensional (5D) compressed sensing (CS) framework for simultaneous Cartesian single (SQ) and triple quantum (TQ) sodium imaging of in vivo human brain at 3.0 and 7.0 T. 3D 23 Na MQC MRI requires multi-echo paired with phase-cycling and exhibits thus a multidimensional space. A joint reconstruction framework to exploit the sparsity in all imaging dimensions by extending the conventional 3D CS framework to 5D was developed. 3D MQC images of simulated brain, phantom and healthy brain volunteers obtained from 3.0 T and 7.0 T were retrospectively and prospectively undersampled. Performance of the CS models were analyzed by means of structural similarity index (SSIM), root mean squared error (RMSE), signal-to-noise ratio (SNR) and signal quantification of tissue sodium concentration and TQ/SQ ratio. It was shown that an acceleration of three-fold, leading to less than min of scan time with a resolution of at 3.0 T, are possible. 5D CS improved SSIM by 3%, 5%, 1% and reduced RMSE by 50%, 30%, 8% for in vivo SQ, TQ, and TQ/SQ ratio maps, respectively. Furthermore, for the first time prospective undersampling enabled unprecedented high resolution from to MQC images of in vivo human brain at 7.0 T without extending acquisition time. 5D CS proved to allow up to three-fold acceleration retrospectively on 3.0 T data. 2-fold acceleration was demonstrated prospectively at 7.0 T to reach higher spatial resolution of 23 Na MQC MRI.

3D Ultrasonic Brain Imaging with Deep Learning Based on Fully Convolutional Networks.

Compared to magnetic resonance imaging (MRI) and X-ray computed tomography (CT), ultrasound imaging is safer, faster, and more widely applicable. However, the use of conventional ultrasound in transcranial brain imaging for adults is predominantly hindered by the high acoustic impedance contrast between the skull and soft tissue. This study introduces a 3D AI algorithm, Brain Imaging Full Convolution Network (BIFCN), combining waveform modeling and deep learning for precise brain ultrasound reconstruction. We constructed a network comprising one input layer, four convolution layers, and one pooling layer to train our algorithm. In the simulation experiment, the Pearson correlation coefficient between the reconstructed and true images was exceptionally high. In the laboratory, the results showed a slightly lower but still impressive coincidence degree for 3D reconstruction, with pure water serving as the initial model and no prior information required. The 3D network can be trained in 8 h, and 10 samples can be reconstructed in just 12.67 s. The proposed 3D BIFCN algorithm provides a highly accurate and efficient solution for mapping wavefield frequency domain data to 3D brain models, enabling fast and precise brain tissue imaging. Moreover, the frequency shift phenomenon of blood may become a hallmark of BIFCN learning, offering valuable quantitative information for whole-brain blood imaging.

CReg-KD: Model refinement via confidence regularized knowledge distillation for brain imaging.

One of the core challenges of deep learning in medical image analysis is data insufficiency, especially for 3D brain imaging, which may lead to model over-fitting and poor generalization. Regularization strategies such as knowledge distillation are powerful tools to mitigate the issue by penalizing predictive distributions and introducing additional knowledge to reinforce the training process. In this paper, we revisit knowledge distillation as a regularization paradigm by penalizing attentive output distributions and intermediate representations. In particular, we propose a Confidence Regularized Knowledge Distillation (CReg-KD) framework, which adaptively transfers knowledge for distillation in light of knowledge confidence. Two strategies are advocated to regularize the global and local dependencies between teacher and student knowledge. In detail, a gated distillation mechanism is proposed to soften the transferred knowledge globally by utilizing the teacher loss as a confidence score. Moreover, the intermediate representations are attentively and locally refined with key semantic context to mimic meaningful features. To demonstrate the superiority of our proposed framework, we evaluated the framework on two brain imaging analysis tasks (i.e. Alzheimer's Disease classification and brain age estimation based on T1-weighted MRI) on the Alzheimer's Disease Neuroimaging Initiative dataset including 902 subjects and a cohort of 3655 subjects from 4 public datasets. Extensive experimental results show that CReg-KD achieves consistent improvements over the baseline teacher model and outperforms other state-of-the-art knowledge distillation approaches, manifesting that CReg-KD as a powerful medical image analysis tool in terms of both promising prediction performance and generalizability.

Inflating 2D convolution weights for efficient generation of 3D medical images

Background and Objective: The generation of three-dimensional (3D) medical images has great application potential since it takes into account the 3D anatomical structure. Two problems prevent effective training of a 3D medical generative model: (1) 3D medical images are expensive to acquire and annotate, resulting in an insufficient number of training images, and (2) a large number of parameters are involved in 3D convolution. Methods: We propose a novel GAN model called 3D Split&Shuffle-GAN. To address the 3D data scarcity issue, we first pre-train a two-dimensional (2D) GAN model using abundant image slices and inflate the 2D convolution weights to improve the initialization of the 3D GAN. Novel 3D network architectures are proposed for both the generator and discriminator of the GAN model to significantly reduce the number of parameters while maintaining the quality of image generation. Several weight inflation strategies and parameter-efficient 3D architectures are investigated. Results: Experiments on both heart (Stanford AIMI Coronary Calcium) and brain (Alzheimer’s Disease Neuroimaging Initiative) datasets show that our method leads to improved 3D image generation quality (14.7 improvements on Frchet inception distance) with significantly fewer parameters (only 48.5% of the baseline method). Conclusions: We built a parameter-efficient 3D medical image generation model. Due to the efficiency and effectiveness, it has the potential to generate high-quality 3D brain and heart images for real use cases.

An efficient memory reserving-and-fading strategy for vector quantization based 3D brain segmentation and tumor extraction using an unsupervised deep learning network.

Deep learning networks are state-of-the-art approaches for 3D brain image segmentation, and the radiological characteristics extracted from tumors are of great significance for clinical diagnosis, treatment planning, and treatment outcome evaluation. However, two problems have been the hindering factors in brain image segmentation techniques. One is that deep learning networks require large amounts of manually annotated data. Another issue is the computational efficiency of 3D deep learning networks. In this study, we propose a vector quantization (VQ)-based 3D segmentation method that employs a novel unsupervised 3D deep embedding clustering (3D-DEC) network and an efficiency memory reserving-and-fading strategy. The VQ-based 3D-DEC network is trained on volume data in an unsupervised manner to avoid manual data annotation. The memory reserving-and-fading strategy beefs up model efficiency greatly. The designed methodology makes deep learning-based model feasible for biomedical image segmentation. The experiment is divided into two parts. First, we extensively evaluate the effectiveness and robustness of the proposed model on two authoritative MRI brain tumor databases (i.e., IBSR and BrainWeb). Second, we validate the model using real 3D brain tumor data collected from our institute for clinical practice significance. Results show that our method (without data manual annotation) has superior accuracy ( Tanimoto coefficient on IBSR, 97.5% TP and 97.7% TN on BrainWeb, and 91% Dice, 88% sensitivity and 87% specificity on real brain data) and remarkable efficiency (speedup ratio is 18.72 on IBSR, 31.16 on BrainWeb, 31.00 on real brain data) compared to the state-of-the-art methods. The results show that our proposed model can address the lacks of manual annotations, and greatly increase computation speedup with competitive segmentation accuracy compared to other state-of-the-art 3D CNN models. Moreover, the proposed model can be used for tumor treatment follow-ups every 6months, providing critical details for surgical and postoperative treatment by correctly extracting numerical radiomic features of tumors.

Efficient Training on Alzheimer's Disease Diagnosis with Learnable Weighted Pooling for 3D PET Brain Image Classification.

Three-dimensional convolutional neural networks (3D CNNs) have been widely applied to analyze Alzheimer's disease (AD) brain images for a better understanding of the disease progress or predicting the conversion from cognitively impaired (CU) or mild cognitive impairment status. It is well-known that training 3D-CNN is computationally expensive and with the potential of overfitting due to the small sample size available in the medical imaging field. Here we proposed a novel 3D-2D approach by converting a 3D brain image to a 2D fused image using a Learnable Weighted Pooling (LWP) method to improve efficient training and maintain comparable model performance. By the 3D-to-2D conversion, the proposed model can easily forward the fused 2D image through a pre-trained 2D model while achieving better performance over different 3D and 2D baselines. In the implementation, we chose to use ResNet34 for feature extraction as it outperformed other 2D CNN backbones. We further showed that the weights of the slices are location-dependent and the model performance relies on the 3D-to-2D fusion view, with the best outcomes from the coronal view. With the new approach, we were able to reduce 75% of the training time and increase the accuracy to 0.88, compared with conventional 3D CNNs, for classifying amyloid-beta PET imaging from the AD patients from the CU participants using the publicly available Alzheimer's Disease Neuroimaging Initiative dataset. The novel 3D-2D model may have profound implications for timely AD diagnosis in clinical settings in the future.

Evaluation of Enhanced 3D Brain Image Using Ultrashort TE Sequence with Inversion Recovery for Preoperative Examination of Brain Tumor: Phantom Study Compared with MPRAGE

It is important to obtain accurate anatomical information with little distortion in preoperative examination of brain tumors. Using PETRA, which is an ultrashort echo time (UTE) sequence that is less affected by magnetic susceptibility artifacts, we determined the optimal imaging conditions (radial views [RV] and inversion time [TI]) for IR-PETRA using the inversion recovery (IR) method and compared it with MPRAGE. IR-PETRA was found to be slightly inferior to MPRAGE in sharpness under the optimum conditions (RV=100,000 and TI=500 ms), but it was significantly improved by using a high RV value, and SNR and CNR were higher than or equal to MPRAGE. It is suggested that IR-PETRA may be an alternative sequence of MPRAGE in preoperative examination of brain tumors.

Maternal organophosphate flame retardant exposure alters the developing mesencephalic dopamine system in fetal rat.

Organophosphate flame retardants (OPFRs) have become the predominant substitution for legacy brominated flame retardants but there is concern about their potential developmental neurotoxicity (DNT). OPFRs readily dissociate from the fireproofed substrate to the environment, and they (or their metabolites) have been detected in diverse matrices including air, water, soil, and biota, including human urine and breastmilk. Given this ubiquitous contamination, it becomes increasingly important to understand the potential effects of OPFRs on the developing nervous system. We have previously shown that maternal exposure to OPFRs results in neuroendocrine disruption, alterations to developmental metabolism of serotonin (5-HT) and axonal extension in male fetal rats, and potentiates adult anxiety-like behaviors. The development of the serotonin and dopamine systems occur in parallel and interact, therefore, we first sought to enhance our prior 5-HT work by first examining the ascending 5-HT system on embryonic day 14 using whole mount clearing of fetal heads and 3-dimensional (3D) brain imaging. We also investigated the effects of maternal OPFR exposure on the development of the mesocortical dopamine system in the same animals through 2-dimensional and 3D analysis following immunohistochemistry for tyrosine hydroxylase (TH). Maternal OPFR exposure induced morphological changes to the putative ventral tegmental area and substantia nigra in both sexes and reduced the overall volume of this structure in males, whereas 5-HT nuclei were unchanged. Additionally, dopaminergic axogenesis was disrupted in OPFR exposed animals, as the dorsoventral spread of ventral telencephalic TH afferents were greater at embryonic day 14, while sparing 5-HT fibers. These results indicate maternal exposure to OPFRs alters the development trajectory of the embryonic dopaminergic system and adds to growing evidence of OPFR DNT.

Attention 3D U-Net with Multiple Skip Connections for Segmentation of Brain Tumor Images.

Among researchers using traditional and new machine learning and deep learning techniques, 2D medical image segmentation models are popular. Additionally, 3D volumetric data recently became more accessible, as a result of the high number of studies conducted in recent years regarding the creation of 3D volumes. Using these 3D data, researchers have begun conducting research on creating 3D segmentation models, such as brain tumor segmentation and classification. Since a higher number of crucial features can be extracted using 3D data than 2D data, 3D brain tumor detection models have increased in popularity among researchers. Until now, various significant research works have focused on the 3D version of the U-Net and other popular models, such as 3D U-Net and V-Net, while doing superior research works. In this study, we used 3D brain image data and created a new architecture based on a 3D U-Net model that uses multiple skip connections with cost-efficient pretrained 3D MobileNetV2 blocks and attention modules. These pretrained MobileNetV2 blocks assist our architecture by providing smaller parameters to maintain operable model size in terms of our computational capability and help the model to converge faster. We added additional skip connections between the encoder and decoder blocks to ease the exchange of extracted features between the two blocks, which resulted in the maximum use of the features. We also used attention modules to filter out irrelevant features coming through the skip connections and, thus, preserved more computational power while achieving improved accuracy.

Phase-based fast 3D high-resolution quantitative T2 MRI in 7 T human brain imaging

Magnetic resonance imaging (MRI) is a powerful and versatile technique that offers a range of physiological, diagnostic, structural, and functional measurements. One of the most widely used basic contrasts in MRI diagnostics is transverse relaxation time (T2)-weighted imaging, but it provides only qualitative information. Realizing quantitative high-resolution T2 mapping is imperative for the development of personalized medicine, as it can enable the characterization of diseases progression. While ultra-high-field (≥ 7 T) MRI offers the means to gain new insights by increasing the spatial resolution, implementing fast quantitative T2 mapping cannot be achieved without overcoming the increased power deposition and radio frequency (RF) field inhomogeneity at ultra-high-fields. A recent study has demonstrated a new phase-based T2 mapping approach based on fast steady-state acquisitions. We extend this new approach to ultra-high field MRI, achieving quantitative high-resolution 3D T2 mapping at 7 T while addressing RF field inhomogeneity and utilizing low flip angle pulses; overcoming two main ultra-high field challenges. The method is based on controlling the coherent transverse magnetization in a steady-state gradient echo acquisition; achieved by utilizing low flip angles, a specific phase increment for the RF pulses, and short repetition times. This approach simultaneously extracts both T2 and RF field maps from the phase of the signal. Prior to in vivo experiments, the method was assessed using a 3D head-shaped phantom that was designed to model the RF field distribution in the brain. Our approach delivers fast 3D whole brain images with submillimeter resolution without requiring special hardware, such as multi-channel transmit coil, thus promoting high usability of the ultra-high field MRI in clinical practice.

Non-rigid multi-modal brain image registration based on two-stage generative adversarial nets

This paper has proposed the two-stage generative adversarial nets (GAN) based multi-modal brain image registration method. The first stage uses the GAN to indirectly recover image deformation and the second stage employs the GAN to directly estimate the registered image. At the first stage, the GAN uses structural representations of reference and float images generated by the improved modality independent neighborhood descriptor (MIND) as inputs, and outputs the predicted structural representation of the registered image. At the second stage, the GAN estimates the intensities of the registered image according to the generated structural representations at the first stage. The two GANs designed based on the U-Net are trained using the different loss functions related to mean square error, mean structural similarity (MSSIM) and the least squares loss. Experiments on the 3D brain images from the Atlas, BrainWeb and RIRE datasets show that compared with state-of-the-art registration methods based on sum of squared differences on entropy images (ESSD), MIND, self-similarity context (SSC), voxelmorph (Morph), volume tweening network (VTN) and adversarial similarity network (ASN), the proposed method has competitive or better registration performance in terms of dice value, MSSIM and computational efficiency.

Deep residual-SVD network for brain image registration

Objective. Medical image registration aims to find the deformation field that can align two images in a spatial position. A medical image registration method based on U-Net architecture has been proposed currently. However, U-Net architecture has few training parameters, which leads to weak learning ability, and it ignores the adverse effects of image noise on the registration accuracy. The article aims at addressing the problem of weak network learning ability and the adverse effects of noisy images on registration. Approach. Here we propose a novel unsupervised 3D brain image registration framework, which introduces the residual unit and singular value decomposition (SVD) denoising layer on the U-Net architecture. Residual unit solves the problem of network degradation, that is, registration accuracy becomes saturated and then degrades rapidly with the increase in network depth. SVD denoising layer uses the estimated model order for SVD-based low-rank image reconstruction. we use Akaike information criterion to estimate the appropriate model order, which is used to remove noise components. We use the exponential linear unit (ELU) as the activation function, which is more robust to noise than other peers. Main results. The proposed method is evaluated on the publicly available brain MRI datasets: Mindboggle101 and LPBA40. Experimental results demonstrate our method outperforms several state-of-the-art methods for the metric of Dice Score. The mean number of folding voxels and registration time are comparable to state-of-the-art methods. Significance. This study shows that Deep Residual-SVD Network can improve registration accuracy. This study also demonstrate that the residual unit can enhance the learning ability of the network, the SVD denoising layer can denoise effectively, and the ELU is more robust to noise.

Remu-Net: Multi-Branch Net Framework for 3D Brain Tumor Segmentation

As one of the fatal human diseases, early detection of brain tumors can effectively save patients’ lives. Brain tumor image segmentation is of great practical importance for physicians to perform brain tumor diagnoses quickly. Due to the data complexity of 3D brain images, it is impractical to segment out tumor regions manually, so automatic and reliable methods can be utilized instead of manual work to achieve accurate segmentation of tumor regions. In this paper, we propose an end-to-end, more efficient brain tumor MRI segmentation model, REMU-Net, for the problems of multi-scale feature extraction and difficulty in small target feature extraction in 3D brain tumor image segmentation. Firstly, design and use the multi-channel parallel M-RepVGG module as a decoder to achieve multi-scale feature fusion. Secondly, embedding dilated convolution with different dilated rates in the DM-RepVGG module of the encoder to better extract features at different scales. Finally, introduce the expectation-maximizing attention in the network to better extract the features of the internal details of the tumor. The experimental results on the BraTS2018 validation dataset are Dice scores of 80.93%, 90.13%, and 86.15%, respectively. Experimental results on the BraTS2019 validation dataset can be achieved with Dice scores of 78.29%, 90.65%, and 82.77%, respectively.