A deep learning–based method for improving reliability of multicenter diffusion kurtosis imaging with varied acquisition protocols

Abstract

Multicenter magnetic resonance imaging is gaining more popularity in large-sample projects. Since both varying hardware and software across different centers cause unavoidable data heterogeneity across centers, its impact on reliability in study outcomes has also drawn much attention recently.One fundamental issue arises in how to derive model parameters reliably from image data of varying quality. This issue is even more challenging for advanced diffusion methods such as diffusion kurtosis imaging (DKI). Recently, deep learning-based methods have been demonstrated with their potential for robust and efficient computation of diffusion-derived measures. Inspired by these approaches, the current study specifically designed a framework based on a three-dimensional hierarchical convolutional neural network, to jointly reconstruct and harmonize DKI measures from multicenter acquisition to reformulate these to a state-of-the-art hardware using data from traveling subjects. The results from the harmonized data acquired with different protocols show that: 1) the inter-scanner variation of DKI measures within white matter was reduced by 51.5% in mean kurtosis, 65.9% in axial kurtosis, 53.7% in radial kurtosis, and 61.5% in kurtosis fractional anisotropy, respectively; 2) data reliability of each single scanner was enhanced and brought to the level of the reference scanner; and 3) the harmonization network was able to reconstruct reliable DKI values from high data variability. Overall the results demonstrate the feasibility of the proposed deep learning-based method for DKI harmonization and help to simplify the protocol setup procedure for multicenter scanners with different hardware and software configurations.

Data harmonisation for information fusion in digital healthcare: A state-of-the-art systematic review, meta-analysis and future research directions

“…Deep harmonics for diffusion kurtosis imaging (Deep HDKI ): Tong et al. [119] carried out a concise architecture with three 3D-convolution layers for diffusion kurtosis images (DKI). The paired data was generated using an iterative technique called linear least square and were non-linearly registered to diffusion-weighted images acquired on the target scanner using the computational tools.…”

“…Given the mean and standard deviation , the coefficient of variation (COV) is defined as , which depicts the degree of variation in respect to the population mean [ 17 , 28 , 47 , 69 , 90 , 103 , 119 , 125 , 153 ]. The Multivariate COV (MCOV) is used to quantify the variability of features between different cohorts, with a lower value indicating better reproducibility [154] .…”

“…Many researchers measured the RMSE between the harmonised samples and the ground truth targets to assess the replicability [ 49 , 57 , 81 , 90 , 91 , 103 , 114 , 115 , 117 , 119 , 120 , 125 , 130 , 131 ]. For instance, Moyer et al.…”

“…Radiomics refers to the extraction and analysis of a large number of quantitative image features from medical images obtained by CT, PET, MRI, and other imaging modalities [164] . In addition to the studies on phantoms [ 61 , 65 , 94 ], research refer to MRI mainly focus on the brain [ 17 , 47 49 , 51 53 , 55 , 57 , 58 , 63 , 66 , 86 , 87 , 90 92 , 95 , 97 , 99 , 100 , 115 , 117 - 119 , 125 , 126 , 129 131 , 143 , 144 , 157 ], breast [ 101 , 111 ], while those refer to CT focus on ear [60] , liver [ 50 , 64 ] and lung [ 34 , 120 ]. In addition to CT, PET and MRI, optical coherence tomography also suffers severe scanner variability, which can be eased by computational data harmonisation approaches [129] .…”

A deep learning-based multisite neuroimage harmonization framework established with traveling-subject dataset

“…Second, brain characteristics are considered independently in these models, largely neglecting the spatial and topological relationships among brain regions. To overcome these defects, recently proposed DL-based harmonization methods, including U-net (Dewey et al, 2019), cycle-generative adversarial network (Modanwal et al, 2020), or three-dimensional convolutional neural network (Tong et al, 2020), allow for mapping the complex abstract representations of the nonlinear spatial pattern of the site effects in MRI data. These models have been primarily applied to the harmonization of diffusion tensor images (Moyer et al, 2020), structural images (Zuo et al, 2021), and morphological measurements (Zhao et al, 2019), successfully eliminating the site effect in such data with complex spatial or topological information.…”