Gadewar, S. | bioRxiv Science

bioRxiv2023

The Influence of Brain MRI Defacing Algorithms on Brain-Age Predictions via 3D Convolutional Neural Networks

In brain imaging research, it is becoming standard practice to remove the face from the individuals 3D structural MRI scan to ensure data privacy standards are met. Face removal - or defacing - is being advocated for large, multi-site studies where data is transferred across geographically diverse sites. Several methods have been developed to limit the loss of important brain data by accurately and precisely removing non-brain facial tissue. At the same time, deep learning methods such as convolutional neural networks (CNNs) are increasingly being used in medical imaging research for diagnostic classification and prognosis in neurological diseases. These neural networks train predictive models based on patterns in large numbers of images. Because of this, defacing scans could remove informative data. Here, we evaluated 4 popular defacing methods to identify the effects of defacing on brain age prediction - a common benchmarking task of predicting a subjects chronological age from their 3D T1-weighted brain MRI. We compared brain-age calculations using defaced MRIs to those that were directly brain extracted, and those with both brain and face. Significant differences were present when comparing average per-subject error rates between algorithms in both the defaced brain data and the extracted facial tissue. Results also indicated brain age accuracy depends on defacing and the choice of algorithm. In a secondary analysis, we also examined how well comparable CNNs could predict chronological age from the facial region only (the extracted portion of the defaced image), as well as visualize areas of importance in facial tissue for predictive tasks using CNNs. We obtained better performance in age prediction when using the extracted face portion alone than images of the brain, suggesting the need for caution when defacing methods are used in medical image analysis.

bioinformatics↗

bioRxiv2022

Style Transfer Generative Adversarial Networks to Harmonize Multi-Site MRI to a Single Reference Image to Avoid Over-Correction

Recent work within neuroimaging consortia have aimed to identify reproducible, and often subtle, brain signatures of psychiatric or neurological conditions. To allow for high-powered brain imaging analyses, it is often necessary to pool MR images that were acquired with different protocols across multiple scanners. Current retrospective harmonization techniques have shown promise in removing cross-site image variation. However, most statistical approaches may over-correct for technical, scanning-related, variation as they cannot distinguish between confounded image-acquisition based variability and cross-site population variability. Such statistical methods often require that datasets contain subjects or patient groups with similar clinical or demographic information to isolate the acquisition-based variability. To overcome this limitation, we consider cross-site MRI image harmonization as a style transfer problem rather than a domain transfer problem. Using a fully unsupervised deep-learning framework based on a generative adversarial network (GAN), we show that MR images can be harmonized by inserting the style information encoded from a single reference image, without knowing their site/scanner labels a priori. We trained our model using data from five large-scale multi-site datasets with varied demographics. Results demonstrated that our style-encoding model can harmonize MR images, and match intensity profiles, without relying on traveling subjects. This model also avoids the need to control for clinical, diagnostic, or demographic information. We highlight the effectiveness of our method for clinical research by comparing extracted cortical and subcortical features, brain-age estimates, and case-control effect sizes before and after the harmonization. We showed that our harmonization removed the cross-site variances, while preserving the anatomical information and clinical meaningful patterns. We further demonstrated that with a diverse training set, our method successfully harmonized MR images collected from unseen scanners and protocols, suggesting a promising novel tool for ongoing collaborative studies. Source code is released in USC-IGC/style_transfer_harmonization (github.com).

neuroscience↗