Search

Background: Since the 2016 WHO guidelines, glioma diagnosis has entered an era of integrated diagnosis, combining tissue pathology and molecular pathology. The WHO has focused on promoting the application of molecular diagnosis in the classification of central nervous system tumors. Genetic information such as IDH1 and 1p/19q are important molecular markers, and pathological grading is also a key clinical indicator. However, obtaining genetic pathology labels is more costly than conventional MRI images, resulting in a large number of missing labels in realistic modeling., Method: We propose a training strategy based on label encoding and a corresponding loss function to enable the model to effectively utilize data with missing labels. Additionally, we integrate a graph model with genes and pathology-related clinical prior knowledge into the ResNet backbone to further improve the efficacy of diagnosis. Ten-fold cross-validation experiments were conducted on a large dataset of 1072 patients., Results: The classification area under the curve (AUC) values are 0.93, 0.91, and 0.90 for IDH1, 1p/19q status, and grade (LGG/HGG), respectively. When the label miss rate reached 59.3 %, the method improved the AUC by 0.09, 0.10, and 0.04 for IDH1, 1p/19q, and pathological grade, respectively, compared to the same backbone without the missing label strategy., Conclusions: Our method effectively utilizes data with missing labels and integrates clinical prior knowledge, resulting in improved diagnostic performance for glioma genetic and pathological markers, even with high rates of missing labels., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 Elsevier Ltd. All rights reserved.)

Objectives: Anti-HER2 targeted therapy significantly reduces risk of relapse in HER2 + breast cancer. New measures are needed for a precise risk stratification to guide (de-)escalation of anti-HER2 strategy., Methods: A total of 726 HER2 + cases who received no/single/dual anti-HER2 targeted therapies were split into three respective cohorts. A deep learning model (DeepTEPP) based on preoperative breast magnetic resonance (MR) was developed. Patients were scored and categorized into low-, moderate-, and high-risk groups. Recurrence-free survival (RFS) was compared in patients with different risk groups according to the anti-HER2 treatment they received, to validate the value of DeepTEPP in predicting treatment efficacy and guiding anti-HER2 strategy., Results: DeepTEPP was capable of risk stratification and guiding anti-HER2 treatment strategy: DeepTEPP-Low patients (60.5%) did not derive significant RFS benefit from trastuzumab (p = 0.144), proposing an anti-HER2 de-escalation. DeepTEPP-Moderate patients (19.8%) significantly benefited from trastuzumab (p = 0.048), but did not obtain additional improvements from pertuzumab (p = 0.125). DeepTEPP-High patients (19.7%) significantly benefited from dual HER2 blockade (p = 0.045), suggesting an anti-HER2 escalation., Conclusions: DeepTEPP represents a pioneering MR-based deep learning model that enables the non-invasive prediction of adjuvant anti-HER2 effectiveness, thereby providing valuable guidance for anti-HER2 (de-)escalation strategies. DeepTEPP provides an important reference for choosing the appropriate individualized treatment in HER2 + breast cancer patients, warranting prospective validation., Clinical Relevance Statement: We built an MR-based deep learning model DeepTEPP, which enables the non-invasive prediction of adjuvant anti-HER2 effectiveness, thus guiding anti-HER2 (de-)escalation strategies in early HER2-positive breast cancer patients., Key Points: • DeepTEPP is able to predict anti-HER2 effectiveness and to guide treatment (de-)escalation. • DeepTEPP demonstrated an impressive prognostic efficacy for recurrence-free survival and overall survival. • To our knowledge, this is one of the very few, also the largest study to test the efficacy of a deep learning model extracted from breast MR images on HER2-positive breast cancer survival and anti-HER2 therapy effectiveness prediction., (© 2024. The Author(s).)

Objectives: As a few types of glioma, young high-risk low-grade gliomas (HRLGGs) have higher requirements for postoperative quality of life. Although adjuvant chemotherapy with delayed radiotherapy is the first treatment strategy for HRLGGs, not all HRLGGs benefit from it. Accurate assessment of chemosensitivity in HRLGGs is vital for making treatment choices. This study developed a multimodal fusion radiomics (MFR) model to support radiochemotherapy decision-making for HRLGGs., Methods: A MFR model combining macroscopic MRI and microscopic pathological images was proposed. Multiscale features including macroscopic tumor structure and microscopic histological layer and nuclear information were grabbed by unique paradigm, respectively. Then, these features were adaptively incorporated into the MFR model through attention mechanism to predict the chemosensitivity of temozolomide (TMZ) by means of objective response rate and progression free survival (PFS)., Results: Macroscopic tumor texture complexity and microscopic nuclear size showed significant statistical differences (p < 0.001) between sensitivity and insensitivity groups. The MFR model achieved stable prediction results, with an area under the curve of 0.950 (95% CI: 0.942-0.958), sensitivity of 0.833 (95% CI: 0.780-0.848), specificity of 0.929 (95% CI: 0.914-0.936), positive predictive value of 0.833 (95% CI: 0.811-0.860), and negative predictive value of 0.929 (95% CI: 0.914-0.934). The predictive efficacy of MFR was significantly higher than that of the reported molecular markers (p < 0.001). MFR was also demonstrated to be a predictor of PFS., Conclusions: A MFR model including radiomics and pathological features predicts accurately the response postoperative TMZ treatment., Clinical Relevance Statement: Our MFR model could identify young high-risk low-grade glioma patients who can have the most benefit from postoperative upfront temozolomide (TMZ) treatment., Key Points: • Multimodal radiomics is proposed to support the radiochemotherapy of glioma. • Some macro and micro image markers related to tumor chemotherapy sensitivity are revealed. • The proposed model surpasses reported molecular markers, with a promising area under the curve (AUC) of 0.95., (© 2023. The Author(s), under exclusive licence to European Society of Radiology.)

The quality of magnetic resonance (MR) images obtained with different instruments and imaging parameters varies greatly. A large number of heterogeneous images are collected, and they suffer from acquisition variation. Such imaging quality differences will have a great impact on the radiomics analysis. The main differences in MR images include modality mismatch (M), intensity distribution variance (I), and layer-spacing differences (L), which are referred to as MIL differences in this paper for convenience. An MIL normalization system is proposed to reconstruct uneven MR images into high-quality data with complete modality, a uniform intensity distribution and consistent layer spacing. Three radiomics tasks, including tumor segmentation, pathological grading and genetic diagnosis of glioma, were used to verify the effect of MIL normalization on radiomics analysis. Three retrospective glioma datasets were analyzed in this study: BraTs (285 cases), TCGA (112 cases) and HuaShan (403 cases). They were used to test the effect of MIL on three different radiomics tasks, including tumor segmentation, pathological grading and genetic diagnosis. MIL normalization included three components: multimodal synthesis based on an encoder-decoder network, intensity normalization based on CycleGAN, and layer-spacing unification based on Statistical Parametric Mapping (SPM). The Dice similarity coefficient, areas under the curve (AUC) and six other indicators were calculated and compared after different normalization steps. The MIL normalization system can improved the Dice coefficient of segmentation by 9% (P < .001), the AUC of pathological grading by 32% (P < .001), and IDH1 status prediction by 25% (P < .001) when compared to non-normalization. The proposed MIL normalization system provides high-quality standardized data, which is a prerequisite for accurate radiomics analysis., (Copyright © 2021 Elsevier Ltd. All rights reserved.)

In magnetic resonance imaging (MRI), different imaging settings lead to various intensity distributions for a specific imaging object, which brings huge diversity to data-driven medical applications. To standardize the intensity distribution of magnetic resonance (MR) images from multiple centers and multiple machines using one model, a cycle generative adversarial network (CycleGAN)-based framework is proposed. It utilizes a unified forward generative adversarial network (GAN) path and multiple independent backward GAN paths to transform images in different groups into a single reference one. To preserve image details and prevent resolution loss, two jump connections are applied in the CycleGAN generators. A weak-pair strategy is designed to fully utilize the prior knowledge of the organ structure and promote the performance of the GANs. The experiments were conducted on a T2-FLAIR image database with 8192 slices from 489 patients. The database was obtained from four hospitals and five MRI scanners and was divided into nine groups with different imaging parameters. Compared with the representative algorithms, the peak signal-to-noise ratio, the histogram correlation, and the structural similarity were increased by 3.7%, 5.1%, and 0.1% on average, respectively; the gradient magnitude similarity deviation, the mean square error, and the average disparity were reduced by 19.0%, 15.7%, and 9.9% on average, respectively. Experiments also showed the robustness of the proposed model with a different training set configuration and effectiveness of the proposed framework over the original CycleGAN. Therefore, the MR images with different imaging settings could be efficiently standardized by the proposed method, which would benefit various data-driven applications.

Background: Frequent somatic mutations of BRAF and CTNNB1 were identified in both histological subtypes of craniopharyngioma (adamantinomatous and papillary) which shed light on target therapy to cure this oncogenic disease. The aim of this study was to investigate the noninvasive MRI-based radiomics diagnosis to detect BRAF and CTNNB1 mutations in craniopharyngioma patients., Methods: Forty-four patients pathologically diagnosed as adamantinomatous craniopharyngioma (ACP) or papillary craniopharyngioma (PCP) were retrospectively studied. High-throughput features were extracted from manually segmented tumors in MR images of each case. The modifications-robustness in region of interests and Random Forest-based feature selection methods were adopted to select the most significant features. Random forest classifier with 10-fold cross-validation was applied to build our radiomics model., Results: Four features were selected to make pathological diagnosis between ACP and PCP with area under the receiver operating characteristic curve (AUC) of 0.89, accurancy (ACC) of 0.86, sensitivity (SENS) of 0.89 and specificity (SPEC) of 0.85. The other two features were applied to estimate BRAF V600E mutation with AUC of 0.91, ACC of 0.93, SENS of 0.83 and SPEC of 0.97. Accurate predication of CTNNB1 mutation by three selected features was realized with AUC of 0.93, ACC of 0.86, SENS of 0.86 and SPEC of 0.86., Conclusions: We developed a reliable MRI-based radiomics approach to perform pathological and molecular diagnosis in craniopharyngioma patients with considerably accurate prediction, which could offer potential guidance for clinical decision-making.

Purpose of the Study: Due to the totally different therapeutic regimens needed for primary central nervous system lymphoma (PCNSL) and glioblastoma (GBM), accurate differentiation of the two diseases by noninvasive imaging techniques is important for clinical decision-making., Materials and Methods: Thirty cases of PCNSL and 66 cases of GBM with conventional T1-contrast magnetic resonance imaging (MRI) were analyzed in this study. Convolutional neural networks was used to segment tumor automatically. A modified scale invariant feature transform (SIFT) method was utilized to extract three-dimensional local voxel arrangement information from segmented tumors. Fisher vector was proposed to normalize the dimension of SIFT features. An improved genetic algorithm (GA) was used to extract SIFT features with PCNSL and GBM discrimination ability. The data-set was divided into a cross-validation cohort and an independent validation cohort by the ratio of 2:1. Support vector machine with the leave-one-out cross-validation based on 20 cases of PCNSL and 44 cases of GBM was employed to build and validate the differentiation model., Results: Among 16,384 high-throughput features, 1356 features show significant differences between PCNSL and GBM with p < 0.05 and 420 features with p < 0.001. A total of 496 features were finally chosen by improved GA algorithm. The proposed method produces PCNSL vs. GBM differentiation with an area under the curve (AUC) curve of 99.1% (98.2%), accuracy 95.3% (90.6%), sensitivity 85.0% (80.0%) and specificity 100% (95.5%) on the cross-validation cohort (and independent validation cohort)., Conclusions: Since the local voxel arrangement characterization provided by SIFT features, proposed method produced more competitive PCNSL and GBM differentiation performance by using conventional MRI than methods based on advanced MRI.

Glioblastoma is the most aggressive malignant brain tumor with poor prognosis. Radiomics is a newly emerging and promising technique to reveal the complex relationships between high-throughput medical image features and deep information of disease including pathology, biomarkers and genomics. An approach was developed to investigate the internal relationship between magnetic resonance imaging (MRI) features and the age-related origins of glioblastomas based on a quantitative radiomics method. A fully automatic image segmentation method was applied to segment the tumor regions from three dimensional MRI images. 555 features were then extracted from the image data. By analyzing large numbers of quantitative image features, some predictive and prognostic information could be obtained by the radiomics approach. 96 patients diagnosed with glioblastoma pathologically have been divided into two age groups (<45 and ≥45 years old). As expected, there are 101 features showing the consistency with the age groups (T test, p < .05), and unsupervised clustering results of those features also show coherence with the age difference (T test, p= .006). In conclusion, glioblastoma in different age groups present different radiomics-feature patterns with statistical significance, which indicates that glioblastoma in different age groups should have different pathologic, protein, or genic origins.

Deep learning-based radiomics (DLR) was developed to extract deep information from multiple modalities of magnetic resonance (MR) images. The performance of DLR for predicting the mutation status of isocitrate dehydrogenase 1 (IDH1) was validated in a dataset of 151 patients with low-grade glioma. A modified convolutional neural network (CNN) structure with 6 convolutional layers and a fully connected layer with 4096 neurons was used to segment tumors. Instead of calculating image features from segmented images, as typically performed for normal radiomics approaches, image features were obtained by normalizing the information of the last convolutional layers of the CNN. Fisher vector was used to encode the CNN features from image slices of different sizes. High-throughput features with dimensionality greater than 1.6*10 4 were obtained from the CNN. Paired t-tests and F-scores were used to select CNN features that were able to discriminate IDH1. With the same dataset, the area under the operating characteristic curve (AUC) of the normal radiomics method was 86% for IDH1 estimation, whereas for DLR the AUC was 92%. The AUC of IDH1 estimation was further improved to 95% using DLR based on multiple-modality MR images. DLR could be a powerful way to extract deep information from medical images.

The segmentation of high-grade gliomas (HGG) using magnetic resonance imaging (MRI) data is clinically meaningful in neurosurgical practice, but a challenging task. Currently, most segmentation methods are supervised learning with labeled training sets. Although these methods work well in most cases, they typically require time-consuming manual labeling and pre-trained models. In this work, we propose an automatically unsupervised segmentation toolbox based on the clustering algorithm and morphological processing, named AUCseg. With our toolbox, the whole tumor was first extracted by clustering on T2-FLAIR images. Then, based on the mask acquired with whole tumor segmentation, the enhancing tumor was segmented on the post-contrast T1-weighted images (T1-CE) using clustering methods. Finally, the necrotic regions were segmented by morphological processing or clustering on T2-weighted images. Compared with K-means, Mini-batch K-means, and Fuzzy C Means (FCM), the Gaussian Mixture Model (GMM) clustering performs the best in our toolbox. We did a multi-sided evaluation of our toolbox in the BraTS2018 dataset and demonstrated that the whole tumor, tumor core, and enhancing tumor can be automatically segmented using default hyper-parameters with Dice score 0.8209, 0.7087, and 0.7254, respectively. The computing time of our toolbox for each case is around 22 seconds, which is at least 3 times faster than other state-of-the-art unsupervised methods. In addition, our toolbox has an option to perform semi-automatic segmentation via manually setup hyper-parameters, which could improve the segmentation performance. Our toolbox, AUCseg, is publicly available on Github. (https://github.com/Haifengtao/AUCseg). [ABSTRACT FROM AUTHOR]

Background: We developed an image patch classification-based method to detect early ischemic stroke. The accuracy of this method was >75%. We aimed to analyze patients' image data to identify interference factors that would affect its accuracy. Methods: We conducted a retrospective analysis of 162 patients who were hospitalized with acute ischemic stroke. Factors related to the noncontrast computed tomography (ncCT) determination results were analyzed according to the patient's sex, age, clinical symptoms, cerebral infarction volume, cerebral infarction location, and whether or not the white matter high (WMH) signal was combined. Results: The volume of cerebral infarction was positively correlated with the predicted results. The correct percentages of patients with volumes >1 and <1 mL were 59.18 and 83.19%, respectively, and the difference was statistically significant (p = 0.001). The correct percentage of the internal capsule region (47.1%) was significantly lower than that of the other groups (p = 0.011). The correct percentage of lateral ventricular paraventricular infarction was significantly lower than that of non-lateral ventricle patients (70.8 vs. 85.7%). In patients with lateral ventricular paraventricular infarction, if the WMH was combined, the correct percentage will decreased further as the Fazekas level increased. The correct percentage of lateral ventricle infarction combined with Fazekas 3 was 40.0%, which was statistically significant compared with the patient having Fazekas 0 with lateral ventricular infarction (p = 0.01). Conclusions: WMH had a similar computed tomography appearance to cerebral infarction and could interfere with the prediction of the cerebral infarction region by ncCT. This result provides a reference for clinicians to choose imaging methods for identifying acute cerebral infarction areas. [ABSTRACT FROM AUTHOR]

With the development of big data, Radiomics and deep-learning methods based on magnetic resonance (MR) images, it is necessary to conduct large databases containing MR images from multiple centers. Having huge intensity distribution differences among images reduced or even eliminated, robust computer-aided diagnosis models could be established. Therefore, an optimized intensity standardization model is proposed. The network structure, loss function, and data input strategy were optimized to better avoid the image resolution loss during transformation. The experimental dataset was obtained from five MR scanners located in four hospitals and was divided into nine groups based on the imaging parameters, during which 9152 MR images from 499 participants were collected. Experiments show the superiority of the proposed method to the previously proposed unified model in resolution metrics including the peak signal-to-noise ratio, structural similarity, visual information fidelity, universal quality index, and image fidelity criterion. Another experiment further shows the advantage of the proposed method in increasing the effectiveness of following computer-aided diagnosis models by better preservation of MR image details. Moreover, the advantage over conventional standardization methods are also shown. Thus, MR images from different centers can be standardized using the proposed method, which will facilitate numerous data-driven medical imaging studies. [ABSTRACT FROM AUTHOR]

As one of the most widely used medical imaging methods, magnetic resonance imaging (MRI) and its corresponding automatic image analysis are research hotspots in computer-aided medical diagnosis. However, due to the disparities in imaging parameters and scanner characteristics, the variation in intensity distribution between scanners results in performance reduction in automatic image analysis and diagnosis. This paper aims at obtaining sub-region intensity distribution and forming a robust non-rigid intensity transforming function in order to standardise the intensities of MR images. The proposed method includes multi-modality image registration, sub-region standard intensity estimation, weighted n -dimensional point cloud generation and global intensity transformation. Between the target images and reference images, this novel method could not only avoid the intensity distortion due to the inconsistency of inter-tissue brightness relationship, but also reduce the dependence on the accuracy of multi-modality MR image registration. Experiments show performance enhancement in peak signal-to-noise ratio, average disparity, histogram correlation, mean square error and structural similarity over the existing methods. Therefore, the intensities of MR images from different scanners could be standardised by the proposed method, so that multi-centre/multi-machine correlation could be promoted. [ABSTRACT FROM AUTHOR]

• A general and concise reinforcement supervision strategy is designed for the segmentation of multi-ROI in medical images. • A generative-adversarial structure with a reused decoder is proposed to implement our strategy, which further simplifies network structure, reduces the training difficulty and improves the inference efficiency. • Three modules including a dilated convolution chain, a fusion block, and a location encoder are proposed to strengthen the segmentation performance of the framework. • In the new version, we carefully revised the manuscript based on the comments, as detailed in the response letter. A new multiple region of interest (multi-ROI) segmentation framework, RsALUNet, is proposed in this paper, whose backbone was U-Net. Rs represented the reinforcement-supervision strategy by utilizing adversarial learning (AL) between U-Net's decoders and an additional discriminator, which was based on the differences among the segmentation results (diff S e g s ) and labels of multi-ROI (diff L a b e l s ). As the AL progressed, diff S e g s was increasingly similar to diff Labels , and this further contributed to a more accurate segmentation of each member in the multi-ROI. In addition to the Rs strategy, three blocks were proposed to enhance RsALUNet, namely, a dilated convolution chain providing diverse and large receptive fields to accommodate different target sizes, a fusion block integrating features of large targets to small ones to optimize the segmentation of the latter, and a location-encoder block extracting multi-scale positional information to enhance the model's attention to the ROI. RsALUNet was evaluated through three multi-ROI segmentation tasks using different imaging modalities, including X-ray, ultrasound, and magnetic-resonance (MR) imaging. The mean Dice coefficient (Dice) increased from 1.1% to 8.5% compared to the other frameworks. The results demonstrate the promising adaptability and extendibility of our strategy and RsALUNet for multi-ROI segmentation in X-ray, ultrasound, and MR images. [ABSTRACT FROM AUTHOR]

Background: Tumor location served as an important prognostic factor in glioma patients was considered to postulate molecular features according to cell origin theory. However, anatomic distribution of unique molecular subtypes was not widely investigated. The relationship between molecular phenotype and histological subgroup were also vague based on tumor location. Our group focuses on the study of glioma anatomic location of distinctive molecular subgroups and histology subtypes, and explores the possibility of their consistency based on clinical background.Methods: We retrospectively reviewed 143 cases with both molecular information (IDH1/TERT/1p19q) and MRI images diagnosed as cerebral diffuse gliomas. The anatomic distribution was analyzed between distinctive molecular subgroups and its relationship with histological subtypes. The influence of tumor location, molecular stratification and histology diagnosis on survival outcome was investigated as well.Results: Anatomic locations of cerebral diffuse glioma indicate varied clinical outcome. Based on that, it can be stratified into five principal molecular subgroups according to IDH1/TERT/1p19q status. Triple-positive (IDH1 and TERT mutation with 1p19q codeletion) glioma tended to be oligodendroglioma present with much better clinical outcome compared to TERT mutation only group who is glioblastoma inclined (median overall survival 39 months VS 18 months). Five molecular subgroups were demonstrated with distinctive locational distribution. This kind of anatomic feature is consistent with its corresponding histological subtypes.Discussion: Each molecular subgroup in glioma has unique anatomic location which indicates distinctive clinical outcome. Molecular diagnosis can be served as perfect complementary tool for the precise diagnosis. Integration of histomolecular diagnosis will be much more helpful in routine clinical practice in the future. [ABSTRACT FROM AUTHOR]

• We propose a R-Transformer network to achieve precise brain tumor segmentation by combining R-Transformer with U-Net. • ERTN employs the ranking attention mechanism, which helps the model improve training efficiency and reduce computation cost. • ERTN can provide higher brain tumor segmentation accuracy than the CNNs or Transformers based methods. Gliomas are the most prevalent and destructive forms of primary brain tumors with a high mortality rate. Magnetic resonance imaging is extensively employed in the examination of gliomas. Segmenting brain glioma on magnetic resonance images has significant clinical value. However, the blur boundary of gliomas, variability in the shape, location, and size make segmentation extremely challenging. In this paper, we propose a new method to segment gliomas from three-dimensional brain magnetic resonance images accurately. We propose an efficient R-Transformer network with dual encoders (ERTN) to achieve precise segmentation by innovatively combining R-Transformer with U-Net. Specifically, ERTN constructs a feature branch and a patch branch, capturing complex semantic features and global context information. Moreover, features generated from the feature branch and patch branch are up-sampled and combined with low- and high-resolution CNN features in the decoder to enable precise localization. At last, ERTN employs the ranking attention mechanism in Transformer (R-Transformer), which helps the model focus on helpful information to improve training efficiency and reduce computation cost. Experiments on the 2017 BRATS dataset prove that ERTN achieves satisfactory performance, with a Dice similarity coefficient of 83.20%, 77.93%, and 72.59% on the whole tumor, tumor core, and enhanced tumor segmentation. For the Hausdorff distance index, we obtained the scores of 5.30, 4.60, and 5.50 for the whole tumor, tumor core, and enhanced tumor, respectively. Our results suggest that ERTN improves the segmentation accuracy and reduces computation cost, which performs better than the existing convolution- and transformer-based methods. [ABSTRACT FROM AUTHOR]

Purpose: Assessment of the fetal cerebellar volume on 3D ultrasound data sets is very important in the clinical evaluation of the fetal growth and health. However, the irregular shape of the cerebellum and the strong artifacts of ultrasound images complicate the segmentation without manual intervention. In this paper, the authors propose an approach to locate the cerebellum automatically, which is considered as a prework of the segmentation. Methods: The authors present a weighted Hough transform and a constrained randomized Hough transform to detect the fetal brain midline and the skull, respectively. By combining the location information of these two structures with local image features, a constrained probabilistic boosting tree is then proposed to search the cerebellum. Results: This algorithm was tested on ultrasound volumes of the fetal head with the gestational age ranging from 20 to 33 weeks. Compared with manual measurements, this algorithm obtained a satisfactory performance with the mean Dice similarity coefficient of 0.92 and the average processing time of 0.75 s per case. Conclusions: The results demonstrate that the proposed method is an automatic, fast, and accurate tool for searching the fetal cerebellum on ultrasound volumes. [ABSTRACT FROM AUTHOR]

• A 3D UNet-based model was proposed to conduct cross-modality synthesis of MR images. • Multi-scale and residual learning framework were developed to improve the effect. • Experimental results on three datasets demonstrate the promising performance. Multi-modality MRI has been increasingly used for disease diagnosis and prognosis evaluation. However, the acquisitions of multi-modal images always face some practical difficulties such as the increased cost and patient discomfort. In this paper, we proposed an improved 3D UNet model for MRI modality synthesis and use the synthetic modalities to improve the performance of the following image analysis tasks such as tumor segmentation and tumor grading. We developed our model with three improvements on the basic 3D UNet model. First, we constructed a coarse-to-fine resolution fusion network to improve the ability of structural texture preservation. Second, we designed a convolutional fusion-based residual learning framework to learn mapping function, which leads to more detail preservation as well as less artifacts in the background. Third, for many-to-one synthesis, to combine the input multi-modalities more effectively, we introduced a new network module to fuse the UNet output feature maps of each modality. One private and two public datasets including a total of 493 cases were used for synthesis validation. Our proposed method surpasses the best existing method about 2.0 dB and 0.8 dB in PSNRs on the two datasets, respectively. Furthermore, two separate radiomics applications were used to demonstrate the value of our modality synthesis method. In the image segmentation application, integrating the modalities synthesized by our method has improved the segmentation Dice by 1.3%. In gliomas grading diagnosis application, integrating the synthesized modalities improved the diagnostic accuracy by 8.3%. [ABSTRACT FROM AUTHOR]