Hippocampal subfields segmentation in brain MR images using generative adversarial networks

The brain is considered a main important part of human body as it participate to control all the activities of human. Under some unhealthy conditions the abnormal and uncontrolled growth of tissue may take place which is unnatural process. This kind of process taking place in human brain is referred as “brain tumor”. The brain tumor may be of cancerous (malignant) or may be of non-cancerous (Benign) and its detection is a very much important to reduced the death rate of humans suffering from tumor. The detection of brain tumor can be performed by using various image processing techniques like brain magnetic resonance imaging (MRI), Computed Tomography (CT), Electroencephalography (EEG), Positron Emission Tomography (PET), Magneto Encephalography (MEG) etc. Among these techniques the brain MRI is widely adopted in the world due to its significant features. But to detect tumor in the brain MRI, the processes like preprocessing, segmentation and classification are must. This paper deals with the various aspects of the brain tumor detection. The paper discusses the significant researches which are meant for brain tumor detection through MRI quality enhancement, MRI segmentation and MRI classification (as cancerous or not). The outcome of the paper discussed that there were no proper researches are performed to detect the brain tumor, focused on good quality of image, less researches on segmentation and classification of brain tumor. Finally the paper gives the future research scope to overcome the addressed research gap.

Hippocampal subfields play important and divergent roles in both memory formation and early diagnosis of many neurological diseases, but automatic subfield segmentation is less explored due to its small size and poor image contrast. In this paper, we propose an automatic learning-based hippocampal subfields segmentation framework using multi-modality 3TMR images, including T1 MRI and resting-state fMRI (rs-fMRI). To do this, we first acquire both 3T and 7T T1 MRIs for each training subject, and then the 7T T1 MRI are linearly registered onto the 3T T1 MRI. Six hippocampal subfields are manually labeled on the aligned 7T T1 MRI, which has the 7T image contrast but sits in the 3T T1 space. Next, corresponding appearance and relationship features from both 3T T1 MRI and rs-fMRI are extracted to train a structured random forest as a multi-label classifier to conduct the segmentation. Finally, the subfield segmentation is further refined iteratively by additional context features and updated relationship features. To our knowledge, this is the first work that addresses the challenging automatic hippocampal subfields segmentation using 3T routine T1 MRI and rs-fMRI. The quantitative comparison between our results and manual ground truth demonstrates the effectiveness of our method. Besides, we also find that (a) multi-modality features significantly improved subfield segmentation performance due to the complementary information among modalities; (b) automatic segmentation results using 3T multimodality images are partially comparable to those on 7T T1 MRI.

The segmentation of magnetic resonance (MR) brain images is an important problem in medical imaging. Accurate segmentation of MR brain images allows a detail study of 3D brain tissue anatomy. It is also of great interest in the study of many brain disorders, where accurate volumetric measurement of the disorders is often required. In view of the importance of the task, much effort has been spent on finding accurate and efficient algorithms for the MRI segmentation problem. This chapter attempts to give the readers an overview of the MR brain segmentation problem, the various image artifacts that are often encountered, and describe some of the current approaches in this area, as well as our own work. To facilitate discussion, we broadly divide current MR brain image segmentation algorithms into three categories: classification-based, region-based, and contour-based methods, and discuss the merits and limitations of these approaches. Following a review of existing methods, we describe our approach for MR brain image segmentation in detail. Our approach is based on a clustering-for-classification framework, using a novel variant of the fuzzy c-means algorithm. We show that by incorporating two key ideas into the clustering algorithm, we are able to take into account the local spatial context, to compensate for the intensity nonuniformity artifact and the partial volume averaging artifact, and to reduce the influence of image noise, during the segmentation process. Extensive experiment results on both simulated and real MR brain images are given to illustrate the effectiveness and robustness of our approach. We conclude this chapter by pointing to some possible future directions in this area.

In present times, automatic computer interpretation of medical images is of great importance for medical diagnosis. Having previously been an obstacle for medical science, early stage diagnoses together with accurately calculating the size and level of disease are very important elements of today's treatment process. The treatment and diagnosis of brain tumors depend on successful processing and evaluation of magnetic resonance (MR) brain images. Image segmentation is the most intractable process, because of the complicated nature of the MR brain images. In processing and analyzing MR images, there is unsuccessful automatic systems or successful semiautomatic system, which is still required large scale experting involvement and assessment. Within this paper the aim is to propose a system that will automatically extract the location, volume and borders of a brain tumor from patient MRI brain images. To achieve this goal, an automated algorithm has been developed, based upon and derived from a successful semi-automatic hybrid algorithm, for automation of the tumor diagnosis process. Patients' MR brain images from varying stages of the diagnosis and treatment were used for testing and validating the automatic algorithm under development. Based upon the achieved results of this algorithm and method, an automatic, sensitive MR brain computer interpretation can be proposed.

Accurate and efficient diagnosis in a short time period is an important part of brain magnetic resonance imaging (MRI) classification. In this paper, we use multi-cluster feature selection (MCFS) method to select efficient features from the primary features for brain MRI classification. The primary features are obtained from a three-level two-dimensional discrete wavelet transform (2D DWT). The selected features are then applied to the K-nearest neighbor (KNN) classifier. We classify the MRI as normal or one of the seven different diseases. The results demonstrate that the proposed method achieves higher accuracy than the other methods in distinguishing different types of disease.

Validation of automatic MRI hippocampal subfield segmentation by histopathological evaluation in patients with temporal lobe epilepsy

BackgroundIn healthy ageing, there is evidence of sex differences in vulnerability of hippocampal subfields to volume loss. However, this has not been investigated in early‐onset Alzheimer’s disease (<65 years; EOAD).MethodWe included 106 subjects: 62 EOAD (A+T+N+, MMSE>15) and 44 healthy controls (HC; A‐T‐N‐) that underwent lumbar puncture for analysis of AD biomarkers, 3T‐MRI scan and neuropsychological assessment. Hippocampal subfield segmentation was performed using T1‐weighted images and Freesurfer 6.0. Volume was adjusted by intracranial volume. Adjusted linear models were used to analyze differences between EOAD and HC and differences between sexes. We calculated Cohen’s d as a measure of the effect size of volume change by sex, restricted to volume differences between EOAD and HC. In EOAD, we used linear models adjusted by age and education to investigate the association of volume loss with 18 cognition z‐scores. Results were adjusted using Bonferroni correction for multiple comparisons.ResultThere were no demographic differences across groups. APOEε4 carriers were higher in EOAD‐female/EOAD‐male than HC‐female. EOAD‐female showed higher T‐Tau and P‐Tau levels than EOAD‐male (all p<0.05; Table1). Comparing EOAD vs. HC, differences were found in volume of all subfields and hippocampus (p<0.05, Bonferroni corrected), except for bilateral parasubiculum and right cornu ammonis (CA) 2/3. No differences were found between EOAD‐female and EOAD‐male (p>0.05). When compared to HC of the same sex, EOAD‐female showed differences in the same regions as the whole sample, while EOAD‐male showed differences in bilateral hippocampal tail and left presubiculum and hippocampus (all p<0.05, Bonferroni corrected, Figure 1). We observed larger effect sizes than in women in these regions, except for left hippocampus (Figure 2).In EOAD, higher volume in left CA2/3 predicted higher memory z‐scores, including free and total learning and delayed total recall (p<0.05 Bonferroni corrected; Figure 3). No associations were found in EOAD‐female nor EOAD‐male.ConclusionIn EOAD, the pattern of volume loss in hippocampal subfields was similar between sexes. However, females showed more marked differences from HC, except for bilateral hippocampal tail and left presubiculum. Further studies are necessary to elucidate the existence of sex differences in EOAD and their implication for cognitive impairment.

Hippocampal subfields, daily stressors, and resting cortisol in individuals at clinical high-risk for psychosis

Performance of semi-automated hippocampal subfield segmentation methods across ages in a pediatric sample

Noise reduction in brain magnetic resonance (MR) images has been a challenging and demanding task. This study develops a new trilateral filter that aims to achieve robust and efficient image restoration. Extended from the bilateral filter, the proposed algorithm contains one additional intensity similarity funct-ion, which compensates for the unique characteristics of noise in brain MR images. An entropy function adaptive to intensity variations is introduced to regulate the contributions of the weighting components. To hasten the computation, parallel computing based on the graphics processing unit (GPU) strategy is explored with emphasis on memory allocations and thread distributions. To automate the filtration, image texture feature analysis associated with machine learning is investigated. Among the 98 candidate features, the sequential forward floating selection scheme is employed to acquire the optimal texture features for regularization. Subsequently, a two-stage classifier that consists of support vector machines and artificial neural networks is established to predict the filter parameters for automation. A speedup gain of 757 was reached to process an entire MR image volume of 256 × 256 × 256 pixels, which completed within 0.5s. Automatic restoration results revealed high accuracy with an ensemble average relative error of 0.53 ± 0.85% in terms of the peak signal-to-noise ratio. This self-regulating trilateral filter outperformed many state-of-the-art noise reduction methods both qualitatively and quantitatively. We believe that this new image restoration algorithm is of potential in many brain MR image processing applications that require expedition and automation.

Every year thousands of patients are diagnosed with a glioma, a type of malignant brain tumor. MRI plays an essential role in the diagnosis and treatment assessment of these patients. Neural networks show great potential to aid physicians in the medical image analysis. This study investigated the creation of synthetic brain T1-weighted (T1), post-contrast T1-weighted (T1CE), T2-weighted (T2), and T2 Fluid Attenuated Inversion Recovery (Flair) MR images. These synthetic MR (synMR) images were assessed quantitatively with four metrics. The synMR images were also assessed qualitatively by an authoring physician with notions that synMR possessed realism in its portrayal of structural boundaries but struggled to accurately depict tumor heterogeneity. Additionally, this study investigated the synMR images created by generative adversarial network (GAN) to overcome the lack of annotated medical image data in training U-Nets to segment enhancing tumor, whole tumor, and tumor core regions on gliomas. Multiple two-dimensional (2D) U-Nets were trained with original BraTS data and differing subsets of the synMR images. Dice similarity coefficient (DSC) was used as the loss function during training as well a quantitative metric. Additionally, Hausdorff Distance 95% CI (HD) was used to judge the quality of the contours created by these U-Nets. The model performance was improved in both DSC and HD when incorporating synMR in the training set. In summary, this study showed the ability to generate high quality Flair, T2, T1, and T1CE synMR images using GAN. Using synMR images showed encouraging results to improve the U-Net segmentation performance and shows potential to address the scarcity of annotated medical images.

The magnetic resonance (MR) imaging has become an indispensable tool for diagnosis and study of various brain diseases. To perform an accurate diagnosis of a brain disease and monitor its evolution and treatment outcomes, a neuroradiologist often needs to measure the volume and assess the changes of shapes in specific brain structures along a series of MR images. In general, brain structures are manually delineated by a radiologist and, therefore, they highly dependend on the professional’s skills. In this study, we proposed the construction of a probabilistic atlas consisting of 3D landmark points automatically detected in a set of MR images. Also, we aimed at investigate its applicability to guide the initial positioning of mesh models based on the deformation of the hippocampus in brain MR images. The normalized Dice Similarity Coefficient (DSC) and the Hausdorff Average Distance (HAD) were used for the quantitative performance evaluation of the proposed method. The results showed that the average values obtained by our atlas-based landmark approach were significantly better (DSC = 0.74/0.70, HAD = 0.70/0.73, for left and right hippocampus, respectively) than our previous initial approaches, such as the template-based landmark (DSC = 0.65/0.61, HAD = 0.88/0.91) and the affine transformation (DSC = 0.58/0.53, HAD = 1.10/1.22).

Neuroscience is of emerging importance along with the contributions of Operational Research to the practices of diagnosing neurodegenerative diseases with computer-aided systems based on brain image analysis. Although multiple biomarkers derived from Magnetic Resonance Imaging (MRI) data have proven to be effective in diagnosing Alzheimer’s disease (AD) and mild cognitive impairment (MCI), no specific system has yet been a part of routine clinical practice. This paper aims to introduce a fully-automated voxel-based procedure, Voxel-MARS, for detection of AD and MCI in early stages of progression. Performance was evaluated on a dataset of 508 MRI volumes gathered from the Alzheimer’s Disease Neuroimaging Initiative database. Data were transformed into a high-dimensional space through a feature extraction process. A novel 3-step feature selection procedure was applied. Multivariate Adaptive Regression Splines method was used as a classifier for the first time in the field of brain MRI analysis. The results were compared to those presented in a previous study on 28 voxel-based methods in terms of their ability to separate control normal (CN) subjects from the ones diagnosed with AD and MCI. It was observed that our method outperformed all of the others in sensitivity (83.58% in AD/CN and 78.38% in MCI/CN classification) with acceptable specificity values (over 85% in both cases). Furthermore, the method worked for discriminating MCI patients which converted to AD in 18 months (MCIc) from non-converters (MCInc) with a sensitivity outcome better than 27 of 28 methods. Overall, it was shown that the proposed method is promising in early detection of AD.

An efficient statistical model tissue classification algorithm is proposed for the segmentation of brain magnetic resonance images. Due to the partial volume effects, many voxels may be composed of a multiple tissue types. To solve this problem, we present an efficient method for brain magnetic resonance images classification. The method uses the Bayesian contextual classifier based on Markov random field models. In the algorithm, each mixture voxels in the MR image is labeled using the maximum a posteriori classifier. A spatial prior is defined based on homogeneous regions while taking into different tissue mixtures. The efficacy of the proposed algorithm is demonstrated by extensive experiments using phantom data.

PurposeMagnetic resonance imaging (MRI)-guided radiotherapy enables adaptive treatment plans based on daily anatomical changes and accurate organ visualization. However, the bias field artifact can compromise image quality, affecting diagnostic accuracy and quantitative analyses. This study aims to assess the impact of bias field correction on 0.35 T pelvis MRIs by evaluating clinical anatomy visualization and generative adversarial network (GAN) auto-segmentation performance.Materials and methods3D simulation MRIs from 60 prostate cancer patients treated on MR-Linac (0.35 T) were collected and preprocessed with the N4ITK algorithm for bias field correction. A 3D GAN architecture was trained, validated, and tested on 40, 10, and 10 patients, respectively, to auto-segment the organs at risk (OARs) rectum and bladder. The GAN was trained and evaluated either with the original or the bias-corrected MRIs. The Dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (HD95th) were computed for the segmented volumes of each patient. The Wilcoxon signed-rank test assessed the statistical difference of the metrics within OARs, both with and without bias field correction. Five radiation oncologists blindly scored 22 randomly chosen patients in terms of overall image quality and visibility of boundaries (prostate, rectum, bladder, seminal vesicles) of the original and bias-corrected MRIs. Bennett’s S score and Fleiss’ kappa were used to assess the pairwise interrater agreement and the interrater agreement among all the observers, respectively.ResultsIn the test set, the GAN trained and evaluated on original and bias-corrected MRIs showed DSC/HD95th of 0.92/5.63 mm and 0.92/5.91 mm for the bladder and 0.84/10.61 mm and 0.83/9.71 mm for the rectum. No statistical differences in the distribution of the evaluation metrics were found neither for the bladder (DSC: p = 0.07; HD95th: p = 0.35) nor for the rectum (DSC: p = 0.32; HD95th: p = 0.63). From the clinical visual grading assessment, the bias-corrected MRI resulted mostly in either no change or an improvement of the image quality and visualization of the organs’ boundaries compared with the original MRI.ConclusionThe bias field correction did not improve the anatomy visualization from a clinical point of view and the OARs’ auto-segmentation outputs generated by the GAN.