Vulnerability of Thalamic Nuclei at CSF Interface During the Entire Course of Multiple Sclerosis.

BACKGROUND AND OBJECTIVES: Thalamic atrophy can be used as a proxy for neurodegeneration in multiple sclerosis (MS). Some data point toward thalamic nuclei that could be affected more than others. However, the dynamic of their changes during MS evolution and the mechanisms driving their differential alterations are still uncertain. METHODS: We paired a large cohort of 1,123 patients with MS with the same number of healthy controls, all scanned with conventional 3D-T1 MRI. To highlight the main atrophic regions at the thalamic nuclei level, we validated a segmentation strategy consisting of deep learning-based synthesis of sequences, which were used for automatic multiatlas segmentation. Then, through a lifespan-based approach, we could model the dynamics of the 4 main thalamic nuclei groups. RESULTS: All analyses converged toward a higher rate of atrophy for the posterior and medial groups compared with the anterior and lateral groups. We also demonstrated that focal MS white matter lesions were associated with atrophy of groups of nuclei when specifically located within the associated thalamocortical projections. The volumes of the most affected posterior group, but also of the anterior group, were better associated with clinical disability than the volume of the whole thalamus. DISCUSSION: These findings point toward the thalamic nuclei adjacent to the third ventricle as more susceptible to neurodegeneration during the entire course of MS through potentiation of disconnection effects by regional factors. Because this information can be obtained even from standard T1-weighted MRI, this paves the way toward such an approach for future monitoring of patients with MS.

Staging of progressive supranuclear palsy-Richardson syndrome using MRI brain charts for the human lifespan.

Brain charts for the human lifespan have been recently proposed to build dynamic models of brain anatomy in normal aging and various neurological conditions. They offer new possibilities to quantify neuroanatomical changes from preclinical stages to death, where longitudinal MRI data are not available. In this study, we used brain charts to model the progression of brain atrophy in progressive supranuclear palsy-Richardson syndrome. We combined multiple datasets (n = 8170 quality controlled MRI of healthy subjects from 22 cohorts covering the entire lifespan, and n = 62 MRI of progressive supranuclear palsy-Richardson syndrome patients from the Four Repeat Tauopathy Neuroimaging Initiative (4RTNI)) to extrapolate lifetime volumetric models of healthy and progressive supranuclear palsy-Richardson syndrome brain structures. We then mapped in time and space the sequential divergence between healthy and progressive supranuclear palsy-Richardson syndrome charts. We found six major consecutive stages of atrophy progression: (i) ventral diencephalon (including subthalamic nuclei, substantia nigra, and red nuclei), (ii) pallidum, (iii) brainstem, striatum and amygdala, (iv) thalamus, (v) frontal lobe, and (vi) occipital lobe. The three structures with the most severe atrophy over time were the thalamus, followed by the pallidum and the brainstem. These results match the neuropathological staging of tauopathy progression in progressive supranuclear palsy-Richardson syndrome, where the pathology is supposed to start in the pallido-nigro-luysian system and spreads rostrally via the striatum and the amygdala to the cerebral cortex, and caudally to the brainstem. This study supports the use of brain charts for the human lifespan to study the progression of neurodegenerative diseases, especially in the absence of specific biomarkers as in PSP.

Lifespan neurodegeneration of the human brain in multiple sclerosis.

Atrophy related to multiple sclerosis (MS) has been found at the early stages of the disease. However, the archetype dynamic trajectories of the neurodegenerative process, even prior to clinical diagnosis, remain unknown. We modeled the volumetric trajectories of brain structures across the entire lifespan using 40,944 subjects (38,295 healthy controls and 2649 MS patients). Then, we estimated the chronological progression of MS by assessing the divergence of lifespan trajectories between normal brain charts and MS brain charts. Chronologically, the first affected structure was the thalamus, then the putamen and the pallidum (around 4 years later), followed by the ventral diencephalon (around 7 years after thalamus) and finally the brainstem (around 9 years after thalamus). To a lesser extent, the anterior cingulate gyrus, insular cortex, occipital pole, caudate and hippocampus were impacted. Finally, the precuneus and accumbens nuclei exhibited a limited atrophy pattern. Subcortical atrophy was more pronounced than cortical atrophy. The thalamus was the most impacted structure with a very early divergence in life. Our experiments showed that lifespan models of most impacted structures could be an important tool for future preclinical/prodromal prognosis and monitoring of MS.

Lifespan Neurodegeneration Of The Human Brain In Multiple Sclerosis.

Background: Atrophy related to Multiple Sclerosis (MS) has been found at the early stages of the disease. However, the archetype dynamic trajectories of the neurodegenerative process, even prior to clinical diagnosis, remain unknown. Methods: We modeled the volumetric trajectories of brain structures across the entire lifespan using 40944 subjects (38295 healthy controls and 2649 MS patients). Then, we estimated the chronological progression of MS by assessing the divergence of lifespan trajectories between normal brain charts and MS brain charts. Results: Chronologically, the first affected structure was the thalamus, then the putamen and the pallidum (3 years later), followed by the ventral diencephalon (7 years after thalamus) and finally the brainstem (9 years after thalamus). To a lesser extent, the anterior cingulate gyrus, insular cortex, occipital pole, caudate and hippocampus were impacted. Finally, the precuneus and accumbens nuclei exhibited a limited atrophy pattern. Conclusion: Subcortical atrophy was more pronounced than cortical atrophy. The thalamus was the most impacted structure with a very early divergence in life. It paves the way toward utilization of these lifespan models for future preclinical/prodromal prognosis and monitoring of MS.

Anatomical MRI staging of frontotemporal dementia variants.

INTRODUCTION: The three clinical variants of frontotemporal dementia (behavioral variant [bvFTD], semantic dementia, and progressive non-fluent aphasia [PNFA]) are likely to develop over decades, from the preclinical stage to death. METHODS: To describe the long-term chronological anatomical progression of FTD variants, we built lifespan brain charts of normal aging and FTD variants by combining 8022 quality-controlled MRIs from multiple large-scale data-bases, including 107 bvFTD, 44 semantic dementia, and 38 PNFA. RESULTS: We report in this manuscript the anatomical MRI staging schemes of the three FTD variants by describing the sequential divergence of volumetric trajectories between normal aging and FTD variants. Subcortical atrophy precedes focal cortical atrophy in specific behavioral and/or language networks, with a "radiological" prodromal phase lasting 8-10 years (time elapsed between the first structural alteration and canonical cortical atrophy). DISCUSSION: Amygdalar and striatal atrophy can be candidate biomarkers for future preclinical/prodromal FTD variants definitions. HIGHLIGHTS: We describe the chronological MRI staging of the most affected structures in the three frontotemporal dementia (FTD) syndromic variants. In behavioral variant of FTD (bvFTD): bilateral amygdalar, striatal, and insular atrophy precedes fronto-temporal atrophy. In semantic dementia: bilateral amygdalar atrophy precedes left temporal and hippocampal atrophy. In progressive non-fluent aphasia (PNFA): left striatal, insular, and thalamic atrophy precedes opercular atrophy.

Midbrain structure volume, estimated myelin and functional connectivity in idiopathic generalised epilepsy.

BACKGROUND: Structural and functional neuroimaging studies often overlook lower basal ganglia structures located in and adjacent to the midbrain due to poor contrast on clinically acquired T1-weighted scans. Here, we acquired T1-weighted, T2-weighted, and resting-state fMRI scans to investigate differences in volume, estimated myelin content and functional connectivity of the substantia nigra (SN), subthalamic nuclei (SubTN) and red nuclei (RN) of the midbrain in IGE. METHODS: Thirty-three patients with IGE (23 refractory, 10 non-refractory) and 39 age and sex-matched healthy controls underwent MR imaging. Midbrain structures were automatically segmented from T2-weighted images and structural volumes were calculated. The estimated myelin content for each structure was determined using a T1-weighted/T2-weighted ratio method. Resting-state functional connectivity analysis of midbrain structures (seed-based) was performed using the CONN toolbox. RESULTS: An increased volume of the right RN was found in IGE and structural volumes of the right SubTN differed between patients with non-refractory and refractory IGE. However, no volume findings survived corrections for multiple comparisons. No myelin alterations of midbrain structures were found for any subject groups. We found functional connectivity alterations including significantly decreased connectivity between the left SN and the thalamus and significantly increased connectivity between the right SubTN and the superior frontal gyrus in IGE. CONCLUSIONS: We report volumetric and functional connectivity alterations of the midbrain in patients with IGE. We postulate that potential increases in structural volumes are due to increased iron deposition that impacts T2-weighted contrast. These findings are consistent with previous studies demonstrating pathophysiological abnormalities of the lower basal ganglia in animal models of generalised epilepsy.

vol2Brain: A New Online Pipeline for Whole Brain MRI Analysis.

Automatic and reliable quantitative tools for MR brain image analysis are a very valuable resource for both clinical and research environments. In the past few years, this field has experienced many advances with successful techniques based on label fusion and more recently deep learning. However, few of them have been specifically designed to provide a dense anatomical labeling at the multiscale level and to deal with brain anatomical alterations such as white matter lesions (WML). In this work, we present a fully automatic pipeline (vol2Brain) for whole brain segmentation and analysis, which densely labels (N > 100) the brain while being robust to the presence of WML. This new pipeline is an evolution of our previous volBrain pipeline that extends significantly the number of regions that can be analyzed. Our proposed method is based on a fast and multiscale multi-atlas label fusion technology with systematic error correction able to provide accurate volumetric information in a few minutes. We have deployed our new pipeline within our platform volBrain (www.volbrain.upv.es), which has been already demonstrated to be an efficient and effective way to share our technology with the users worldwide.

Structural progression of Alzheimer's disease over decades: the MRI staging scheme.

The chronological progression of brain atrophy over decades, from pre-symptomatic to dementia stages, has never been formally depicted in Alzheimer's disease. This is mainly due to the lack of cohorts with long enough MRI follow-ups in cognitively unimpaired young participants at baseline. To describe a spatiotemporal atrophy staging of Alzheimer's disease at the whole-brain level, we built extrapolated lifetime volumetric models of healthy and Alzheimer's disease brain structures by combining multiple large-scale databases (n = 3512 quality controlled MRI from 9 cohorts of subjects covering the entire lifespan, including 415 MRI from ADNI1, ADNI2 and AIBL for Alzheimer's disease patients). Then, we validated dynamic models based on cross-sectional data using external longitudinal data. Finally, we assessed the sequential divergence between normal aging and Alzheimer's disease volumetric trajectories and described the following staging of brain atrophy progression in Alzheimer's disease: (i) hippocampus and amygdala; (ii) middle temporal gyrus; (iii) entorhinal cortex, parahippocampal cortex and other temporal areas; (iv) striatum and thalamus and (v) middle frontal, cingular, parietal, insular cortices and pallidum. We concluded that this MRI scheme of atrophy progression in Alzheimer's disease was close but did not entirely overlap with Braak staging of tauopathy, with a 'reverse chronology' between limbic and entorhinal stages. Alzheimer's disease structural progression may be associated with local tau accumulation but may also be related to axonal degeneration in remote sites and other limbic-predominant associated proteinopathies.

Hippocampal-amygdalo-ventricular atrophy score: Alzheimer disease detection using normative and pathological lifespan models.

In this article, we present an innovative MRI-based method for Alzheimer disease (AD) detection and mild cognitive impairment (MCI) prognostic, using lifespan trajectories of brain structures. After a full screening of the most discriminant structures between AD and normal aging based on MRI volumetric analysis of 3,032 subjects, we propose a novel Hippocampal-Amygdalo-Ventricular Atrophy score (HAVAs) based on normative lifespan models and AD lifespan models. During a validation on three external datasets on 1,039 subjects, our approach showed very accurate detection (AUC ≥ 94%) of patients with AD compared to control subjects and accurate discrimination (AUC = 78%) between progressive MCI and stable MCI (during a 3-year follow-up). Compared to normative modeling, classical machine learning methods and recent state-of-the-art deep learning methods, our method demonstrated better classification performance. Moreover, HAVAs simplicity makes it fully understandable and thus well-suited for clinical practice or future pharmaceutical trials.

A novel deep learning based hippocampus subfield segmentation method.

The automatic assessment of hippocampus volume is an important tool in the study of several neurodegenerative diseases such as Alzheimer's disease. Specifically, the measurement of hippocampus subfields properties is of great interest since it can show earlier pathological changes in the brain. However, segmentation of these subfields is very difficult due to their complex structure and for the need of high-resolution magnetic resonance images manually labeled. In this work, we present a novel pipeline for automatic hippocampus subfield segmentation based on a deeply supervised convolutional neural network. Results of the proposed method are shown for two available hippocampus subfield delineation protocols. The method has been compared to other state-of-the-art methods showing improved results in terms of accuracy and execution time.

Longitudinal detection of new MS lesions using deep learning.

The detection of new multiple sclerosis (MS) lesions is an important marker of the evolution of the disease. The applicability of learning-based methods could automate this task efficiently. However, the lack of annotated longitudinal data with new-appearing lesions is a limiting factor for the training of robust and generalizing models. In this study, we describe a deep-learning-based pipeline addressing the challenging task of detecting and segmenting new MS lesions. First, we propose to use transfer-learning from a model trained on a segmentation task using single time-points. Therefore, we exploit knowledge from an easier task and for which more annotated datasets are available. Second, we propose a data synthesis strategy to generate realistic longitudinal time-points with new lesions using single time-point scans. In this way, we pretrain our detection model on large synthetic annotated datasets. Finally, we use a data-augmentation technique designed to simulate data diversity in MRI. By doing that, we increase the size of the available small annotated longitudinal datasets. Our ablation study showed that each contribution lead to an enhancement of the segmentation accuracy. Using the proposed pipeline, we obtained the best score for the segmentation and the detection of new MS lesions in the MSSEG2 MICCAI challenge.

DeepLesionBrain: Towards a broader deep-learning generalization for multiple sclerosis lesion segmentation.

Recently, segmentation methods based on Convolutional Neural Networks (CNNs) showed promising performance in automatic Multiple Sclerosis (MS) lesions segmentation. These techniques have even outperformed human experts in controlled evaluation conditions such as Longitudinal MS Lesion Segmentation Challenge (ISBI Challenge). However, state-of-the-art approaches trained to perform well on highly-controlled datasets fail to generalize on clinical data from unseen datasets. Instead of proposing another improvement of the segmentation accuracy, we propose a novel method robust to domain shift and performing well on unseen datasets, called DeepLesionBrain (DLB). This generalization property results from three main contributions. First, DLB is based on a large group of compact 3D CNNs. This spatially distributed strategy aims to produce a robust prediction despite the risk of generalization failure of some individual networks. Second, we propose a hierarchical specialization learning (HSL) by pre-training a generic network over the whole brain, before using its weights as initialization to locally specialized networks. By this end, DLB learns both generic features extracted at global image level and specific features extracted at local image level. Finally, DLB includes a new image quality data augmentation to reduce dependency to training data specificity (e.g., acquisition protocol). DLB generalization was validated in cross-dataset experiments on MSSEG'16, ISBI challenge, and in-house datasets. During experiments, DLB showed higher segmentation accuracy, better segmentation consistency and greater generalization performance compared to state-of-the-art methods. Therefore, DLB offers a robust framework well-suited for clinical practice.

Deep learning based MRI contrast synthesis using full volume prediction using full volume prediction.

In Magnetic Resonance Imaging (MRI), depending on the image acquisition settings, a large number of image types or contrasts can be generated showing complementary information of the same imaged subject. This multi-spectral information is highly beneficial since can improve MRI analysis tasks such as segmentation and registration, thanks to pattern ambiguity reduction. However, the acquisition of several contrasts is not always possible due to time limitations and patient comfort constraints. Contrast synthesis has emerged recently as an approximate solution to generate other image types different from those acquired originally. Most of the previously proposed methods for contrast synthesis are slice-based which result in intensity inconsistencies between neighbor slices when applied in 3D. We propose the use of a 3D convolutional neural network (CNN) capable of generating T2 and FLAIR images from a single anatomical T1 source volume. The proposed network is a 3D variant of the UNet that processes the whole volume at once breaking with the inconsistency in the resulting output volumes related to 2D slice or patch-based methods. Since working with a full volume at once has a huge memory demand we have introduced a spatial-to-depth and a reconstruction layer that allows working with the full volume but maintain the required network complexity to solve the problem. Our approach enhances the coherence in the synthesized volume while improving the accuracy thanks to the integrated three-dimensional context-awareness. Finally, the proposed method has been validated with a segmentation method, thus demonstrating its usefulness in a direct and relevant application.

Fully automated delineation of the optic radiation for surgical planning using clinically feasible sequences.

Quadrantanopia caused by inadvertent severing of Meyer's Loop of the optic radiation is a well-recognised complication of temporal lobectomy for conditions such as epilepsy. Dissection studies indicate that the anterior extent of Meyer's Loop varies considerably between individuals. Quantifying this for individual patients is thus an important step to improve the safety profile of temporal lobectomies. Previous attempts to delineate Meyer's Loop using diffusion MRI tractography have had difficulty estimating its full anterior extent, required manual ROI placement, and/or relied on advanced diffusion sequences that cannot be acquired routinely in most clinics. Here we present CONSULT: a pipeline that can delineate the optic radiation from raw DICOM data in a completely automated way via a combination of robust pre-processing, segmentation, and alignment stages, plus simple improvements that bolster the efficiency and reliability of standard tractography. We tested CONSULT on 696 scans of predominantly healthy participants (539 unique brains), including both advanced acquisitions and simpler acquisitions that could be acquired in clinically acceptable timeframes. Delineations completed without error in 99.4% of the scans. The distance between Meyer's Loop and the temporal pole closely matched both averages and ranges reported in dissection studies for all tested sequences. Median scan-rescan error of this distance was 1 mm. When tested on two participants with considerable pathology, delineations were successful and realistic. Through this, we demonstrate not only how to identify Meyer's Loop with clinically feasible sequences, but also that this can be achieved without fundamental changes to tractography algorithms or complex post-processing methods.

Distinct Hippocampal Subfields Atrophy in Older People With Vascular Brain Injuries.

BACKGROUND AND PURPOSE: Many neurological or psychiatric diseases affect the hippocampus during aging. The study of hippocampal regional vulnerability may provide important insights into the pathophysiological mechanisms underlying these processes; however, little is known about the specific impact of vascular brain damage on hippocampal subfields atrophy. METHODS: To analyze the effect of vascular injuries independently of other pathological conditions, we studied a population-based cohort of nondemented older adults, after the exclusion of people who were diagnosed with neurodegenerative diseases during the 14-year clinical follow-up period. Using an automated segmentation pipeline, 1.5T-magnetic resonance imaging at inclusion and 4 years later were assessed to measure both white matter hyperintensities and hippocampal subfields volume. Annualized rates of white matter hyperintensity progression and annualized rates of hippocampal subfields atrophy were then estimated in each participant. RESULTS: We included 249 participants in our analyses (58% women, mean age 71.8, median Mini-Mental State Evaluation 29). The volume of the subiculum at baseline was the only hippocampal subfield volume associated with total, deep/subcortical, and periventricular white matter hyperintensity volumes, independently of demographic variables and vascular risk factors (ß=-0.17, P=0.011; ß=-0.25, P=0.020 and ß=-0.14, P=0.029, respectively). In longitudinal measures, the annualized rate of subiculum atrophy was significantly higher in people with the highest rate of deep/subcortical white matter hyperintensity progression, independently of confounding factors (ß=-0.32, P=0.014). CONCLUSIONS: These cross-sectional and longitudinal findings highlight the links between vascular brain injuries and a differential vulnerability of the subiculum within the hippocampal loop, unbiased of the effect of neurodegenerative diseases, and particularly when vascular injuries affect deep/subcortical structures.

Toward a unified analysis of cerebellum maturation and aging across the entire lifespan: A MRI analysis.

Previous literature about the structural characterization of the human cerebellum is related to the context of a specific pathology or focused in a restricted age range. In fact, studies about the cerebellum maturation across the lifespan are scarce and most of them considered the cerebellum as a whole without investigating each lobule. This lack of study can be explained by the lack of both accurate segmentation methods and data availability. Fortunately, during the last years, several cerebellum segmentation methods have been developed and many databases comprising subjects of different ages have been made publically available. This fact opens an opportunity window to obtain a more extensive analysis of the cerebellum maturation and aging. In this study, we have used a recent state-of-the-art cerebellum segmentation method called CERES and a large data set (N = 2,831 images) from healthy controls covering the entire lifespan to provide a model for 12 cerebellum structures (i.e., lobules I-II, III, IV, VI, Crus I, Crus II, VIIB, VIIIA, VIIIB, IX, and X). We found that lobules have generally an evolution that follows a trajectory composed by a fast growth and a slow degeneration having sometimes a plateau for absolute volumes, and a decreasing tendency (faster in early ages) for normalized volumes. Special consideration is dedicated to Crus II, where slow degeneration appears to stabilize in elder ages for absolute volumes, and to lobule X, which does not present any fast growth during childhood in absolute volumes and shows a slow growth for normalized volumes.

Multi-scale graph-based grading for Alzheimer's disease prediction.

The prediction of subjects with mild cognitive impairment (MCI) who will progress to Alzheimer's disease (AD) is clinically relevant, and may above all have a significant impact on accelerating the development of new treatments. In this paper, we present a new MRI-based biomarker that enables us to accurately predict conversion of MCI subjects to AD. In order to better capture the AD signature, we introduce two main contributions. First, we present a new graph-based grading framework to combine inter-subject similarity features and intra-subject variability features. This framework involves patch-based grading of anatomical structures and graph-based modeling of structure alteration relationships. Second, we propose an innovative multiscale brain analysis to capture alterations caused by AD at different anatomical levels. Based on a cascade of classifiers, this multiscale approach enables the analysis of alterations of whole brain structures and hippocampus subfields at the same time. During our experiments using the ADNI-1 dataset, the proposed multiscale graph-based grading method obtained an area under the curve (AUC) of 81% to predict conversion of MCI subjects to AD within three years. Moreover, when combined with cognitive scores, the proposed method obtained 85% of AUC. These results are competitive in comparison to state-of-the-art methods evaluated on the same dataset.

RegQCNET: Deep quality control for image-to-template brain MRI affine registration.

Affine registration of one or several brain image(s) onto a common reference space is a necessary prerequisite for many image processing tasks, such as brain segmentation or functional analysis. Manual assessment of registration quality is a tedious and time-consuming task, especially in studies comprising a large amount of data. Automated and reliable quality control (QC) becomes mandatory. Moreover, the computation time of the QC must be also compatible with the processing of massive datasets. Therefore, automated deep neural network approaches have emerged as a method of choice to automatically assess registration quality. In the current study, a compact 3D convolutional neural network, referred to as RegQCNET, is introduced to quantitatively predict the amplitude of an affine registration mismatch between a registered image and a reference template. This quantitative estimation of registration error is expressed using the metric unit system. Therefore, a meaningful task-specific threshold can be manually or automatically defined in order to distinguish between usable and non-usable images. The robustness of the proposed RegQCNET is first analyzed on lifespan brain images undergoing various simulated spatial transformations and intensity variations between training and testing. Secondly, the potential of RegQCNET to classify images as usable or non-usable is evaluated using both manual and automatic thresholds. During our experiments, automatic thresholds are estimated using several computer-assisted classification models (logistic regression, support vector machine, Naive Bayes and random forest) through cross-validation. To this end we use an expert's visual QC estimated on a lifespan cohort of 3953 brains. Finally, the RegQCNET accuracy is compared to usual image features such as image correlation coefficient and mutual information. The results show that the proposed deep learning QC is robust, fast and accurate at estimating affine registration error in the processing pipeline.

Author Correction: Multimodal hippocampal subfield grading for Alzheimer's disease classification.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

AssemblyNet: A large ensemble of CNNs for 3D whole brain MRI segmentation.

Whole brain segmentation of fine-grained structures using deep learning (DL) is a very challenging task since the number of anatomical labels is very high compared to the number of available training images. To address this problem, previous DL methods proposed to use a single convolution neural network (CNN) or few independent CNNs. In this paper, we present a novel ensemble method based on a large number of CNNs processing different overlapping brain areas. Inspired by parliamentary decision-making systems, we propose a framework called AssemblyNet, made of two "assemblies" of U-Nets. Such a parliamentary system is capable of dealing with complex decisions, unseen problem and reaching a relevant consensus. AssemblyNet introduces sharing of knowledge among neighboring U-Nets, an "amendment" procedure made by the second assembly at higher-resolution to refine the decision taken by the first one, and a final decision obtained by majority voting. During our validation, AssemblyNet showed competitive performance compared to state-of-the-art methods such as U-Net, Joint label fusion and SLANT. Moreover, we investigated the scan-rescan consistency and the robustness to disease effects of our method. These experiences demonstrated the reliability of AssemblyNet. Finally, we showed the interest of using semi-supervised learning to improve the performance of our method.