The role of noise in denoising models for anomaly detection in medical images.

Pathological brain lesions exhibit diverse appearance in brain images, in terms of intensity, texture, shape, size, and location. Comprehensive sets of data and annotations are difficult to acquire. Therefore, unsupervised anomaly detection approaches have been proposed using only normal data for training, with the aim of detecting outlier anomalous voxels at test time. Denoising methods, for instance classical denoising autoencoders (DAEs) and more recently emerging diffusion models, are a promising approach, however naive application of pixelwise noise leads to poor anomaly detection performance. We show that optimization of the spatial resolution and magnitude of the noise improves the performance of different model training regimes, with similar noise parameter adjustments giving good performance for both DAEs and diffusion models. Visual inspection of the reconstructions suggests that the training noise influences the trade-off between the extent of the detail that is reconstructed and the extent of erasure of anomalies, both of which contribute to better anomaly detection performance. We validate our findings on two real-world datasets (tumor detection in brain MRI and hemorrhage/ischemia/tumor detection in brain CT), showing good detection on diverse anomaly appearances. Overall, we find that a DAE trained with coarse noise is a fast and simple method that gives state-of-the-art accuracy. Diffusion models applied to anomaly detection are as yet in their infancy and provide a promising avenue for further research. Code for our DAE model and coarse noise is provided at: https://github.com/AntanasKascenas/DenoisingAE.

A controlled comparison of thickness, volume and surface areas from multiple cortical parcellation packages.

BACKGROUND: Cortical parcellation is an essential neuroimaging tool for identifying and characterizing morphometric and connectivity brain changes occurring with age and disease. A variety of software packages have been developed for parcellating the brain's cortical surface into a variable number of regions but interpackage differences can undermine reproducibility. Using a ground truth dataset (Edinburgh_NIH10), we investigated such differences for grey matter thickness (GMth), grey matter volume (GMvol) and white matter surface area (WMsa) for the superior frontal gyrus (SFG), supramarginal gyrus (SMG), and cingulate gyrus (CG) from 4 parcellation protocols as implemented in the FreeSurfer, BrainSuite, and BrainGyrusMapping (BGM) software packages. RESULTS: Corresponding gyral definitions and morphometry approaches were not identical across the packages. As expected, there were differences in the bordering landmarks of each gyrus as well as in the manner in which variability was addressed. Rostral and caudal SFG and SMG boundaries differed, and in the event of a double CG occurrence, its upper fold was not always addressed. This led to a knock-on effect that was visible at the neighbouring gyri (e.g., knock-on effect at the SFG following CG definition) as well as gyral morphometric measurements of the affected gyri. Statistical analysis showed that the most consistent approaches were FreeSurfer's Desikan-Killiany-Tourville (DKT) protocol for GMth and BrainGyrusMapping for GMvol. Package consistency varied for WMsa, depending on the region of interest. CONCLUSIONS: Given the significance and implications that a parcellation protocol will have on the classification, and sometimes treatment, of subjects, it is essential to select the protocol which accurately represents their regions of interest and corresponding morphometrics, while embracing cortical variability.

Manually-parcellated gyral data accounting for all known anatomical variability.

Morphometric brain changes occur throughout the lifetime and are often investigated to understand healthy ageing and disease, to identify novel biomarkers, and to classify patient groups. Yet, to accurately characterise such changes, an accurate parcellation of the brain must be achieved. Here, we present a manually-parcellated dataset of the superior frontal, the supramarginal, and the cingulate gyri of 10 healthy middle-aged subjects along with a fully detailed protocol based on two anatomical atlases. Gyral parcels were hand-drawn then reviewed by specialists blinded from the protocol to ensure consistency. Importantly, we follow a procedure that allows accounting for anatomical variability beyond what is usually achieved by standard analysis packages and avoids mutually referring to neighbouring gyri when defining gyral edges. We also provide grey matter thickness, grey matter volume, and white matter surface area information for each parcel. This dataset and corresponding measurements are useful in assessing the accuracy of equivalent parcels and metrics generated by image analysis tools and their impact on morphometric studies.

Do 2-year changes in superior frontal gyrus and global brain atrophy affect cognition?

INTRODUCTION: Metabolic alterations to the superior frontal gyrus (SFG) have been linked to cognitive decline. Whether these indicate structural atrophy, which could be screened for at a larger scale using noninvasive structural imaging, is unknown. METHODS: We assessed annual structural magnetic resonance imaging scans and cognitive data from 3 consecutive years from 204 participants from the AD Neuroimaging Initiative database (mean age 72.24 [8.175] years). We evaluated associations between brain structural changes and performance in the Montreal Cognitive Assessment, Everyday Cognition Visuospatial subtest (ECog Visuospatial), and Functional Assessment Questionnaire. RESULTS: Changes in the surface area of the SFG were associated with changes in the outcome of the ECog Visuospatial test (P < .05), but an inconsistent pattern of association was found between the 2-year global brain atrophy progression and changes in the outcome from the three cognitive tests selected. DISCUSSION: The extent into which (and if) changes in the SFG influence cognition warrant further evaluation in a larger period in more heterogeneous population.

A critical analysis of neuroanatomical software protocols reveals clinically relevant differences in parcellation schemes.

A high replicability in region-of-interest (ROI) morphometric or ROI-based connectivity analyses is essential for such methods to provide biomarkers of good health or disease. In this article, we focus on package design, and more specifically on cortical parcellation protocols, for novel insight into their contribution to inter-package differences. A critical analysis of cortical parcellation protocols from FreeSurfer, BrainSuite, BrainVISA and BrainGyrusMapping revealed major limitations. Details of reference populations are generally missing, cortical variability is not always explicitly accounted for and, more importantly, definition of gyral borders can be inconsistent. We recommend that in the package selection process end users incorporate protocol suitability for the ROIs under investigation, with these particular points in mind, as inter-package differences are likely to be significant and the source of incompatibility between studies' results.

Imaging signatures of meningioma and low-grade glioma: a diffusion tensor, magnetization transfer and quantitative longitudinal relaxation time MRI study.

Differentiation of cerebral tumor pathology currently relies on interpretation of conventional structural MRI and in some cases histology. However, more advanced MRI methods may provide further insight into the organization of cerebral tumors and have the potential to aid diagnosis. The objective of this study was to use multimodal quantitative MRI to measure the imaging signatures of meningioma and low-grade glioma (LGG). Nine adults with meningioma and 11 with LGG were identified, and underwent standard structural, quantitative longitudinal relaxation time (T1) mapping, magnetization transfer and diffusion tensor MRI. Maps of mean (〈D〉), axial (λAX) and radial (λRAD) diffusivity, fractional anisotropy (FA), magnetization transfer ratio (MTR) and T1 were generated on a voxel-by-voxel basis. Using structural and echo-planar T2-weighted MRI, manual region-of-interest segmentation of brain tumor, edema, ipsilateral and contralateral normal-appearing white matter (NAWM) was performed. Differences in imaging signatures between the different tissue types, both absolute mean values and ratios relative to contralateral NAWM, were assessed using t-tests with statistical significance set at p<0.05. For both absolute mean values and ratios relative to contralateral NAWM, there were significant differences in 〈D〉, λAX, λRAD, FA, MTR and T1 between meningioma and LGG tumor tissue, respectively. Only T1 and FA differed significantly between edematous tissue associated with the two tumor types. These results suggest that multimodal MRI biomarkers are significantly different, particularly in tumor tissue, between meningioma and LGG. By using quantitative multimodal MRI it may be possible to identify tumor pathology non-invasively.

Permutation and parametric tests for effect sizes in voxel-based morphometry of gray matter volume in brain structural MRI.

Permutation testing has been widely implemented in voxel-based morphometry (VBM) tools. However, this type of non-parametric inference has yet to be thoroughly compared with traditional parametric inference in VBM studies of brain structure. Here we compare both types of inference and investigate what influence the number of permutations in permutation testing has on results in an exemplar study of how gray matter proportion changes with age in a group of working age adults. High resolution T1-weighted volume scans were acquired from 80 healthy adults aged 25-64years. Using a validated VBM procedure and voxel-based permutation testing for Pearson product-moment coefficient, the effect sizes of changes in gray matter proportion with age were assessed using traditional parametric and permutation testing inference with 100, 500, 1000, 5000, 10000 and 20000 permutations. The statistical significance was set at P<0.05 and false discovery rate (FDR) was used to correct for multiple comparisons. Clusters of voxels with statistically significant (PFDR<0.05) declines in gray matter proportion with age identified with permutation testing inference (N≈6000) were approximately twice the size of those identified with parametric inference (N=3221voxels). Permutation testing with 10000 (N=6251voxels) and 20000 (N=6233voxels) permutations produced clusters that were generally consistent with each other. However, with 1000 permutations there were approximately 20% more statistically significant voxels (N=7117voxels) than with ≥10000 permutations. Permutation testing inference may provide a more sensitive method than traditional parametric inference for identifying age-related differences in gray matter proportion. Based on the results reported here, at least 10000 permutations should be used in future univariate VBM studies investigating age related changes in gray matter to avoid potential false findings. Additional studies using permutation testing in large imaging databanks are required to address the impact of model complexity, multivariate analysis, number of observations, sampling bias and data quality on the accuracy with which subtle differences in brain structure associated with normal aging can be identified.