Automatic segmentation of the choroid plexuses: Method and validation in controls and patients with multiple sclerosis

Abstract

Choroid Plexuses (ChP) are structures located in the ventricles that produce the cerebrospinal fluid (CSF) in the central nervous system. They are also a key component of the blood-CSF barrier. Recent studies have described clinically relevant ChP volumetric changes in several neurological diseases including Alzheimer's, Parkinson's disease, and multiple sclerosis (MS). Therefore, a reliable and automated tool for ChP segmentation on images derived from magnetic resonance imaging (MRI) is a crucial need for large studies attempting to elucidate their role in neurological disorders. Here, we propose a novel automatic method for ChP segmentation in large imaging datasets. The approach is based on a 2-step 3D U-Net to keep preprocessing steps to a minimum for ease of use and to lower memory requirements. The models are trained and validated on a first research cohort including people with MS and healthy subjects. A second validation is also performed on a cohort of pre-symptomatic MS patients having acquired MRIs in routine clinical practice. Our method reaches an average Dice coefficient of 0.72 ± 0.01 with the ground truth and a volume correlation of 0.86 on the first cohort while outperforming FreeSurfer and FastSurfer-based ChP segmentations. On the dataset originating from clinical practice, the method reaches a Dice coefficient of 0.67 ± 0.01 (being close to the inter-rater agreement of 0.64 ± 0.02) and a volume correlation of 0.84. These results demonstrate that this is a suitable and robust method for the segmentation of the ChP both on research and clinical datasets.

Keywords: Choroid plexus; Deep learning; Multiple sclerosis; Radiologically isolated syndrome; Segmentation.

Fig. 1

3D U-Net Architecture and 2-step method. Top row: Architecture of the 3D U-Net used in the architecture. Two bottom rows: description of the 2-step method. Middle row: description of step1, «High resolution» corresponds to images with a dimension of 176 × 240 × 256 with voxels of size 1 mm3; «Low resolution» corresponds to images down-sampled to a dimension of 72 × 96 × 104. Step1 output is thresholded at 0.1 for visualization purposes. (2 column fitting image).

Fig. 2

Examples of manual and automatic segmentations on the testing sets from dataset1 and dataset2. Illustrative subjects with the highest, median, and lowest Dice are presented from top to bottom in each dataset box. Each line corresponds to one subject. The manual segmentation of annotators 1 and 2 are respectively in green and yellow. The prediction of the 2-step method trained with data augmentation is presented in red, thresholded at 0.5, and denoted as “prediction”. The reported Dice is between the prediction and the manual segmentation. When there are two annotators, the Dice shown is computed between the prediction and the 1^st annotator. (2 column fitting image). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3

Results on the independent testing set of dataset1 and dataset2. Lower and upper whiskers represent the minimum and maximum values observed in the dataset. Boxes are bound by the first quartile at the bottom and the third quartile at the top with the center line representing the median of the distribution. (2 column fitting image).