Intensity inhomogeneity correction of MRI images using InhomoNet.

Intensity inhomogeneity is one of the major artifacts in magnetic resonance imaging (MRI). Bias field present in MRI images alters true pixel value and produces spurious varying pixel intensities. This artifact affects the diagnosis by radiologists in a detrimental manner and also degrades the performance of computer-aided diagnosis algorithms such as segmentation. The present work proposes a novel network called InhomoNet for intensity inhomogeneity correction of MRI image. The generator architecture of InhomoNet consists of a new multi-scale local information module at each encoder block that helps to capture features at multiple scales. The horizontal and vertical kernels help to reduce the problems like loss of neighborhood information, gridding issues caused due to large dilated convolution operations. The attention-driven skip connections in the generator network are utilized to transfer optimal semantic and spatial localization information from the encoder to decoder blocks. Further, the present work proposes two new losses functions, i.e. histogram correlation and 3D pixel loss. These losses help to realize pixel consistency across different regions of brain MRI. The inculcation of the L1 loss provides guidance to the upsampling process as it compares the prediction from each decoder block with the ground truth. The proposed method is evaluated on simulated and real MRI data. The comparative analysis with popular state-of-the-art methods depicts the ability of the proposed method to perform intensity inhomogeneity correction accurately.

Intensity Inhomogeneity Correction of Structural MR Images: A Data-Driven Approach to Define Input Algorithm Parameters.

Intensity non-uniformity (INU) in magnetic resonance (MR) imaging is a major issue when conducting analyses of brain structural properties. An inaccurate INU correction may result in qualitative and quantitative misinterpretations. Several INU correction methods exist, whose performance largely depend on the specific parameter settings that need to be chosen by the user. Here we addressed the question of how to select the best input parameters for a specific INU correction algorithm. Our investigation was based on the INU correction algorithm implemented in SPM, but this can be in principle extended to any other algorithm requiring the selection of input parameters. We conducted a comprehensive comparison of indirect metrics for the assessment of INU correction performance, namely the coefficient of variation of white matter (CVWM), the coefficient of variation of gray matter (CVGM), and the coefficient of joint variation between white matter and gray matter (CJV). Using simulated MR data, we observed the CJV to be more accurate than CVWM and CVGM, provided that the noise level in the INU-corrected image was controlled by means of spatial smoothing. Based on the CJV, we developed a data-driven approach for selecting INU correction parameters, which could effectively work on actual MR images. To this end, we implemented an enhanced procedure for the definition of white and gray matter masks, based on which the CJV was calculated. Our approach was validated using actual T1-weighted images collected with 1.5 T, 3 T, and 7 T MR scanners. We found that our procedure can reliably assist the selection of valid INU correction algorithm parameters, thereby contributing to an enhanced inhomogeneity correction in MR images.

Quantitative Evaluation of Intensity Inhomogeneity Correction Methods for Structural MR Brain Images.

The correction of intensity non-uniformity (INU) in magnetic resonance (MR) images is extremely important to ensure both within-subject and across-subject reliability. Here we tackled the problem of objectively comparing INU correction techniques for T1-weighted images, which are the most commonly used in structural brain imaging. We focused our investigations on the methods integrated in widely used software packages for MR data analysis: FreeSurfer, BrainVoyager, SPM and FSL. We used simulated data to assess the INU fields reconstructed by those methods for controlled inhomogeneity magnitudes and noise levels. For each method, we evaluated a wide range of input parameters and defined an enhanced configuration associated with best reconstruction performance. By comparing enhanced and default configurations, we found that the former often provide much more accurate results. Accordingly, we used enhanced configurations for a more objective comparison between methods. For different levels of INU magnitude and noise, SPM and FSL, which integrate INU correction with brain segmentation, generally outperformed FreeSurfer and BrainVoyager, whose methods are exclusively dedicated to INU correction. Nonetheless, accurate INU field reconstructions can be obtained with FreeSurfer on images with low noise and with BrainVoyager for slow and smooth inhomogeneity profiles. Our study may prove helpful for an accurate selection of the INU correction method to be used based on the characteristics of actual MR data.

Influence of signal intensity non-uniformity on brain volumetry using an atlas-based method.

OBJECTIVE: Many studies have reported pre-processing effects for brain volumetry; however, no study has investigated whether non-parametric non-uniform intensity normalization (N3) correction processing results in reduced system dependency when using an atlas-based method. To address this shortcoming, the present study assessed whether N3 correction processing provides reduced system dependency in atlas-based volumetry. MATERIALS AND METHODS: Contiguous sagittal T1-weighted images of the brain were obtained from 21 healthy participants, by using five magnetic resonance protocols. After image preprocessing using the Statistical Parametric Mapping 5 software, we measured the structural volume of the segmented images with the WFU-PickAtlas software. We applied six different bias-correction levels (Regularization 10, Regularization 0.0001, Regularization 0, Regularization 10 with N3, Regularization 0.0001 with N3, and Regularization 0 with N3) to each set of images. The structural volume change ratio (%) was defined as the change ratio (%) = (100 × [measured volume - mean volume of five magnetic resonance protocols] / mean volume of five magnetic resonance protocols) for each bias-correction level. RESULTS: A low change ratio was synonymous with lower system dependency. The results showed that the images with the N3 correction had a lower change ratio compared with those without the N3 correction. CONCLUSION: The present study is the first atlas-based volumetry study to show that the precision of atlas-based volumetry improves when using N3-corrected images. Therefore, correction for signal intensity non-uniformity is strongly advised for multi-scanner or multi-site imaging trials.

Region-based Active Contour Model based on Markov Random Field to Segment Images with Intensity Non-Uniformity and Noise.

This paper represents a new region-based active contour model that can be used to segment images with intensity non-uniformity and high-level noise. The main idea of our proposed method is to use Gaussian distributions with different means and variances with incorporation of intensity non-uniformity model for image segmentation. In order to integrate the spatial information between neighboring pixels in our proposed method, we use Markov Random Field. Our experiments on synthetic images and cerebral magnetic resonance images show the advantages of the proposed method over state-of-art methods, i.e. local Gaussian distribution fitting.

Influence of Signal Intensity Non-Uniformity on Brain Volumetry Using an Atlas-Based Method

OBJECTIVE: Many studies have reported pre-processing effects for brain volumetry; however, no study has investigated whether non-parametric non-uniform intensity normalization (N3) correction processing results in reduced system dependency when using an atlas-based method. To address this shortcoming, the present study assessed whether N3 correction processing provides reduced system dependency in atlas-based volumetry. MATERIALS AND METHODS: Contiguous sagittal T1-weighted images of the brain were obtained from 21 healthy participants, by using five magnetic resonance protocols. After image preprocessing using the Statistical Parametric Mapping 5 software, we measured the structural volume of the segmented images with the WFU-PickAtlas software. We applied six different bias-correction levels (Regularization 10, Regularization 0.0001, Regularization 0, Regularization 10 with N3, Regularization 0.0001 with N3, and Regularization 0 with N3) to each set of images. The structural volume change ratio (%) was defined as the change ratio (%) = (100 x [measured volume - mean volume of five magnetic resonance protocols] / mean volume of five magnetic resonance protocols) for each bias-correction level. RESULTS: A low change ratio was synonymous with lower system dependency. The results showed that the images with the N3 correction had a lower change ratio compared with those without the N3 correction. CONCLUSION: The present study is the first atlas-based volumetry study to show that the precision of atlas-based volumetry improves when using N3-corrected images. Therefore, correction for signal intensity non-uniformity is strongly advised for multi-scanner or multi-site imaging trials.

COMPRESSED SENSING BASED INTENSITY NON-UNIFORMITY CORRECTION.

We present a compressed sensing based approach to remove gain field from magnetic resonance (MR) images of the human brain. During image acquisition, the inhomogeneity present in the radio-frequency (RF) coil appears as shading artifact in the intensity image. The inhomogeneity poses problem in any automatic algorithm that uses intensity as a feature. It has been shown that at low field strength, the shading can be assumed to be a smooth field that is composed of low frequency components. Thus most inhomogeneity correction algorithms assume some kind of explicit smoothness criteria on the field. This sometimes limits the performance of the algorithms if the actual inhomogeneity is not smooth, which is the case at higher field strength. We describe a model-free, non-parametric patch-based approach that uses compressed sensing for the correction. We show that these features enable our algorithm to perform comparably with a current state of the art method N3 on images acquired at low field, while outperforming N3 when the image has non-smooth inhomogeneity, such as 7T images.