Toward MR protocol-agnostic, bias-corrected brain age predicted from clinical-grade MRIs

Abstract

The predicted brain age minus the chronological age ('brain-PAD') could become a clinical biomarker. However, most brain age methods were developed to use research-grade high-resolution T1-weighted MRIs, limiting their applicability to clinical-grade MRIs from multiple protocols. To overcome this, we adopted a double transfer learning approach to develop a brain age model agnostic to modality, resolution, or slice orientation. Using 6,224 clinical MRIs among 7 modalities, scanned from 1,540 patients using 8 scanners among 15 + facilities of the University of Florida's Health System, we retrained a convolutional neural network (CNN) to predict brain age from synthetic research-grade magnetization-prepared rapid gradient-echo MRIs (MPRAGEs) generated by a deep learning-trained 'super-resolution' method. We also modeled the "regression dilution bias", a typical overestimation of younger ages and underestimation of older ages, which correction is paramount for personalized brain age-based biomarkers. This bias was independent of modality or scanner and generalizable to new samples, allowing us to add a bias-correction layer to the CNN. The mean absolute error in test samples was 4.67-6.47 years across modalities, with similar accuracy between original MPRAGEs and their synthetic counterparts. Brain-PAD was also reliable across modalities. We demonstrate the feasibility of clinical-grade brain age predictions, contributing to personalized medicine.

Keywords: Clinical Multimodal MRI; DeepBrainNet; Synthetic MPRAGE; brain age gap; brain-PAD; research-grade MRI; transfer learning.

Figure 1

Distribution of chronological ages in the sample.

Figure 2

Predicted MPRAGEs for three of the five modalities included in this study. The data belongs to a single participant. The top row shows selected slices in the slice, phase-encoding and readout directions (left, right top and right bottom, respectively). The bottom row shows the bottommost (z = −27 mm, slice 46), topmost (z = 53 mm, slice 126), and three intermediate slices (z = −7, 13, and 33mm, slices 66, 86 and 106, respectively) of the 80 slices used for brain age prediction. To show images in world space, trilinear interpolation was used. FLAIR: Fluid attenuated inversion recovery.

Figure 3

Brain age prediction using the selected retrained CNN (VGG16-based DeepBrainNet) for the MRIs of the testing held-out sample. A) Uncorrected and B) bias corrected. The colored lines represent the slope of the linear relation between the chronological and the predicted brain age for each modality. MPRAGE: Magnetization-prepared rapid gradient-echo. T1w: T1-weighted. T2w: T2-weighted. FLAIR: Fluid attenuated inversion recovery. GRE: Gradient Echo. IR: Inversion Recovery. [Modality]-SR: Super-resolution synthetic MPRAGE version of [modality].

Figure 4

Variability of the corrected predicted corrected brain-PADs for each participant (ordered from the lowest to the highest brain-PAD). The bars represent the range of PAD values within each subject. To illustrate the possible causes of some of the outliers, we show the synthetic MPRAGEs of the two images with corrected brain-PAD farthest from the mean in their own within-subject group. These images, of very poor quality, were not detected by the QC described in this study. The inset to the right also shows the original MPRAGE and Synthetic MPRAGEs for the subject having the highest average corrected brain-PAD (26.7 years) in the test set (this participant had no T1wFLAIR or T1w but had two IRs).

Figure 5

Flowchart of the definition of the data domains and the different stages of training and evaluation. After selecting the whole brain MRIs, preprocessing, and performing QC, the final dataset was split for training the CNN model (training set), estimating the parameters of model of the linear bias correction in a different independent sample (the linear correction set) and testing accuracy (external validation in the held-out testing set) and reliability. In more detail, three-fold cross-validation was used on the training set to determine the optimal combination of CNN model, tune hyperparameters and select a linear model. The final training was performed on the training set using the optimal configuration. The trained CNN model was then used to predict the uncorrected brain ages that, together with the chronological ages, were used to estimate the parameters of the linear bias in the independent linear correction set. Finally, the linear model was added on top of the CNN for deployment and tested in the held-out testing set.