Validation of deep learning techniques for quality augmentation in diffusion MRI for clinical studies

The objective of this study is to evaluate the efficacy of deep learning (DL) techniques in improving the quality of diffusion MRI (dMRI) data in clinical applications. The study aims to determine whether the use of artificial intelligence (AI) methods in medical images may result in the loss of critical clinical information and/or the appearance of false information. To assess this, the focus was on the angular resolution of dMRI and a clinical trial was conducted on migraine, specifically between episodic and chronic migraine patients. The number of gradient directions had an impact on white matter analysis results, with statistically significant differences between groups being drastically reduced when using 21 gradient directions instead of the original 61. Fourteen teams from different institutions were tasked to use DL to enhance three diffusion metrics (FA, AD and MD) calculated from data acquired with 21 gradient directions and a b-value of 1000 s/mm2. The goal was to produce results that were comparable to those calculated from 61 gradient directions. The results were evaluated using both standard image quality metrics and Tract-Based Spatial Statistics (TBSS) to compare episodic and chronic migraine patients. The study results suggest that while most DL techniques improved the ability to detect statistical differences between groups, they also led to an increase in false positive. The results showed that there was a constant growth rate of false positives linearly proportional to the new true positives, which highlights the risk of generalization of AI-based tasks when assessing diverse clinical cohorts and training using data from a single group. The methods also showed divergent performance when replicating the original distribution of the data and some exhibited significant bias. In conclusion, extreme caution should be exercised when using AI methods for harmonization or synthesis in clinical studies when processing heterogeneous data in clinical studies, as important information may be altered, even when global metrics such as structural similarity or peak signal-to-noise ratio appear to suggest otherwise.

Viability of AMURA biomarkers from single-shell diffusion MRI in clinical studies

Diffusion Tensor Imaging (DTI) is the most employed method to assess white matter properties using quantitative parameters derived from diffusion MRI, but it presents known limitations that restrict the evaluation of complex structures. The objective of this study was to validate the reliability and robustness of complementary diffusion measures extracted with a novel approach, Apparent Measures Using Reduced Acquisitions (AMURA), with a typical diffusion MRI acquisition from a clinical context in comparison with DTI with application to clinical studies. Fifty healthy controls, 51 episodic migraine and 56 chronic migraine patients underwent single-shell diffusion MRI. Four DTI-based and eight AMURA-based parameters were compared between groups with tract-based spatial statistics to establish reference results. On the other hand, following a region-based analysis, the measures were assessed for multiple subsamples with diverse reduced sample sizes and their stability was evaluated with the coefficient of quartile variation. To assess the discrimination power of the diffusion measures, we repeated the statistical comparisons with a region-based analysis employing reduced sample sizes with diverse subsets, decreasing 10 subjects per group for consecutive reductions, and using 5,001 different random subsamples. For each sample size, the stability of the diffusion descriptors was evaluated with the coefficient of quartile variation. AMURA measures showed a greater number of statistically significant differences in the reference comparisons between episodic migraine patients and controls compared to DTI. In contrast, a higher number of differences was found with DTI parameters compared to AMURA in the comparisons between both migraine groups. Regarding the assessments reducing the sample size, the AMURA parameters showed a more stable behavior than DTI, showing a lower decrease for each reduced sample size or a higher number of regions with significant differences. However, most AMURA parameters showed lower stability in relation to higher coefficient of quartile variation values than the DTI descriptors, although two AMURA measures showed similar values to DTI. For the synthetic signals, there were AMURA measures with similar quantification to DTI, while other showed similar behavior. These findings suggest that AMURA presents favorable characteristics to identify differences of specific microstructural properties between clinical groups in regions with complex fiber architecture and lower dependency on the sample size or assessing technique than DTI.

Anisotropy measure from three diffusion-encoding gradient directions

We propose a method that can provide information about the anisotropy and orientation of diffusion in the brain from only 3 orthogonal gradient directions without imposing additional assumptions. The method is based on the Diffusion Anisotropy (DiA) that measures the distance from a diffusion signal to its isotropic equivalent. The original formulation based on a Spherical Harmonics basis allows to go down to only 3 orthogonal directions in order to estimate the measure. In addition, an alternative simplification and a color-coding representation are also proposed. Acquisitions from a publicly available database are used to test the viability of the proposal. The DiA succeeded in providing anisotropy information from the white matter using only 3 diffusion-encoding directions. The price to pay for such reduced acquisition is an increment in the variability of the data and a subestimation of the metric on those tracts not aligned with the acquired directions. Nevertheless, the calculation of anisotropy information from DMRI is feasible using fewer than 6 gradient directions by using DiA. The method is totally compatible with existing acquisition protocols, and it may provide complementary information about orientation in fast diffusion acquisitions.

Intensity normalization methods in brain FDG-PET quantification

BACKGROUND

The lack of standardization of intensity normalization methods and its unknown effect on the quantification output is recognized as a major drawback for the harmonization of brain FDG-PET quantification protocols. The aim of this work is the ground truth-based evaluation of different intensity normalization methods on brain FDG-PET quantification output.

METHODS

Realistic FDG-PET images were generated using Monte Carlo simulation from activity and attenuation maps directly derived from 25 healthy subjects (adding theoretical relative hypometabolisms on 6 regions of interest and for 5 hypometabolism levels). Single-subject statistical parametric mapping (SPM) was applied to compare each simulated FDG-PET image with a healthy database after intensity normalization based on reference regions methods such as the brain stem (RRBS), cerebellum (RRC) and the temporal lobe contralateral to the lesion (RRTL), and data-driven methods, such as proportional scaling (PS), histogram-based method (HN) and iterative versions of both methods (iPS and iHN). The performance of these methods was evaluated in terms of the recovery of the introduced theoretical hypometabolic pattern and the appearance of unspecific hypometabolic and hypermetabolic findings.

RESULTS

Detected hypometabolic patterns had significantly lower volumes than the introduced hypometabolisms for all intensity normalization methods particularly for slighter reductions in metabolism . Among the intensity normalization methods, RRC and HN provided the largest recovered hypometabolic volumes, while the RRBS showed the smallest recovery. In general, data-driven methods overcame reference regions and among them, the iterative methods overcame the non-iterative ones. Unspecific hypermetabolic volumes were similar for all methods, with the exception of PS, where it became a major limitation (up to 250 cm³) for extended and intense hypometabolism. On the other hand, unspecific hypometabolism was similar far all methods, and usually solved with appropriate clustering.

CONCLUSIONS

Our findings showed that the inappropriate use of intensity normalization methods can provide remarkable bias in the detected hypometabolism and it represents a serious concern in terms of false positives. Based on our findings, we recommend the use of histogram-based intensity normalization methods. Reference region methods performance was equivalent to data-driven methods only when the selected reference region is large and stable.