Efficient Brain Age Prediction from 3D MRI Volumes Using 2D Projections

Encoding 3D information in 2D feature maps for brain CT-Angiography

We propose learnable 3D pooling (L3P), a CNN module designed to compress 3D information into 2D feature maps using anisotropic convolutions and unidirectional max pooling. Specifically, we used L3P followed by a 2D network to generate predictions from 3D brain CT-Angiography (CTA) in the context of large vessel occlusion (LVO). To further demonstrate its versatility, we extended its application to 3D brain MRI analysis for brain age prediction. First, we designed an experiment to classify the LVO-affected hemisphere (left or right), projecting the input CTA into the sagittal plane, which allowed to assess the ability of L3P to encode the 3D location where the location information was in the 3D-to-2D compression axis. Second, we evaluated the use of L3P on LVO detection as a binary classification (presence or absence). We compared the L3P models performance to that of 2D and stroke-specific 3D models. L3P models achieved results equivalent to stroke-specific 3D models while requiring fewer parameters and resources and provided better results than 2D models using maximum intensity projection images as input. The generalizability of L3P approach was evaluated on the LVO-affected hemisphere detection using data from a single site for training/validation and data from 36 other sites for testing, achieving an AUC of 0.83 on the test set. L3P also performed comparably or better than a fully 3D network on a brain age prediction task with a separate T1 MRI dataset, demonstrating its versatility across different tasks and imaging modalities. Additionally, L3P models generated more interpretable feature maps.

Extensive T1-weighted MRI preprocessing improves generalizability of deep brain age prediction models

Brain age is an estimate of chronological age obtained from T1-weighted magnetic resonance images (T1w MRI), representing a straightforward diagnostic biomarker of brain aging and associated diseases. While the current best accuracy of brain age predictions on T1w MRIs of healthy subjects ranges from two to three years, comparing results across studies is challenging due to differences in the datasets, T1w preprocessing pipelines, and evaluation protocols used. This paper investigates the impact of T1w image preprocessing on the performance of four deep learning brain age models from recent literature. Four preprocessing pipelines, which differed in terms of registration transform, grayscale correction, and software implementation, were evaluated. The results showed that the choice of software or preprocessing steps could significantly affect the prediction error, with a maximum increase of 0.75 years in mean absolute error (MAE) for the same model and dataset. While grayscale correction had no significant impact on MAE, using affine rather than rigid registration to brain atlas statistically significantly improved MAE. Models trained on 3D images with isotropic 1mm3 resolution exhibited less sensitivity to the T1w preprocessing variations compared to 2D models or those trained on downsampled 3D images. Our findings indicate that extensive T1w preprocessing improves MAE, especially when predicting on a new dataset. This runs counter to prevailing research literature, which suggests that models trained on minimally preprocessed T1w scans are better suited for age predictions on MRIs from unseen scanners. We demonstrate that, irrespective of the model or T1w preprocessing used during training, applying some form of offset correction is essential to enable the model's performance to generalize effectively on datasets from unseen sites, regardless of whether they have undergone the same or different T1w preprocessing as the training set.

Brain Age Prediction Using 2D Projections Based on Higher-Order Statistical Moments and Eigenslices from 3D Magnetic Resonance Imaging Volumes

Brain age prediction from 3D MRI volumes using deep learning has recently become a popular research topic, as brain age has been shown to be an important biomarker. Training deep networks can be very computationally demanding for large datasets like the U.K. Biobank (currently 29,035 subjects). In our previous work, it was demonstrated that using a few 2D projections (mean and standard deviation along three axes) instead of each full 3D volume leads to much faster training at the cost of a reduction in prediction accuracy. Here, we investigated if another set of 2D projections, based on higher-order statistical central moments and eigenslices, leads to a higher accuracy. Our results show that higher-order moments do not lead to a higher accuracy, but that eigenslices provide a small improvement. We also show that an ensemble of such models provides further improvement.