3D Segmentation of Perivascular Spaces on T1-Weighted 3 Tesla MR Images With a Convolutional Autoencoder and a U-Shaped Neural Network

Abstract

We implemented a deep learning (DL) algorithm for the 3-dimensional segmentation of perivascular spaces (PVSs) in deep white matter (DWM) and basal ganglia (BG). This algorithm is based on an autoencoder and a U-shaped network (U-net), and was trained and tested using T1-weighted magnetic resonance imaging (MRI) data from a large database of 1,832 healthy young adults. An important feature of this approach is the ability to learn from relatively sparse data, which gives the present algorithm a major advantage over other DL algorithms. Here, we trained the algorithm with 40 T1-weighted MRI datasets in which all “visible” PVSs were manually annotated by an experienced operator. After learning, performance was assessed using another set of 10 MRI scans from the same database in which PVSs were also traced by the same operator and were checked by consensus with another experienced operator. The Sorensen-Dice coefficients for PVS voxel detection in DWM (resp. BG) were 0.51 (resp. 0.66), and 0.64 (resp. 0.71) for PVS cluster detection (volume threshold of 0.5 within a range of 0 to 1). Dice values above 0.90 could be reached for detecting PVSs larger than 10 mm3 and 0.95 for PVSs larger than 15 mm3. We then applied the trained algorithm to the rest of the database (1,782 individuals). The individual PVS load provided by the algorithm showed a high agreement with a semi-quantitative visual rating done by an independent expert rater, both for DWM and for BG. Finally, we applied the trained algorithm to an age-matched sample from another MRI database acquired using a different scanner. We obtained a very similar distribution of PVS load, demonstrating the interoperability of this algorithm.

Physiology and Clinical Relevance of Enlarged Perivascular Spaces in the Aging Brain

“…While semiautomated ePVS assessment methods can be helpful tools for clinicians, more fully automated methods that only require 1 or 2 MRI sequences and minimal manual coding are being developed. Our group and others 60 are applying deep learning methods to segment ePVS and acquire volume and count measures throughout the entire brain. Such methods require 50 manually coded images as a training dataset, but once a model is successfully trained and tested, ePVS volume and count measures can be generated in a matter of hours.…”

A critical guide to the automated quantification of perivascular spaces in magnetic resonance imaging

“…To extend this work, it is necessary to explore methods of automated segmentation with lower quality datasets that are more accessible than 7T images. Boutinaud et al (2021) employed another CNN called the u-net with an autoencoder to segment 3T, T1 images ( Boutinaud et al, 2021 ). Typically, model parameters are initialized randomly.…”

“…Trained on 40 manually labeled images, the model achieved a voxel-wise Dice score of 51% in the white matter and 66% in the basal ganglia. Notably, for PVS clusters larger than 10 mm 3 , Dice scores above 90% could be reliably achieved ( Boutinaud et al, 2021 ).…”

“…However, these studies are based on algorithms with certain disadvantages. The conventional methods relying on image processing techniques require further parameter optimization for different datasets ( Ballerini et al, 2016 ; Smith et al, 2020 , Preprint; Boutinaud et al, 2021 ; Bernal et al, 2022 ). Currently, only one fully automatic segmentation pipeline is freely available, making it difficult to replicate previous methods ( Boutinaud et al, 2021 ).…”

Assessment of PVS Filtering Methods Using a Three-Dimensional Computational Model

“…Machine learning, convolutional neural networks specifically, has been utilised in recent years as an alternative to classical techniques leveraging explicit geometrical information (Boutinaud et al, 2021;Dubost et al, 2020Dubost et al, , 2019bDubost et al, , 2019aHuang et al, 2021;Jung et al, 2019;Lian et al, 2018;Yang et al, 2021;Zong et al, 2021). Its use could certainly be advantageous as these techniques can make use of a plethora of visual cues, including intensity, shape, and size, that would enable them to separate PVS from other lesions (Valdes Hernandez et al, 2013).…”