Automated Segmentation of Midbrain Structures in High-Resolution Susceptibility Maps Based on Convolutional Neural Network and Transfer Learning

Abstract

BackgroundAccurate delineation of the midbrain nuclei, the red nucleus (RN), substantia nigra (SN) and subthalamic nucleus (STN), is important in neuroimaging studies of neurodegenerative and other diseases. This study aims to segment midbrain structures in high-resolution susceptibility maps using a method based on a convolutional neural network (CNN).MethodsThe susceptibility maps of 75 subjects were acquired with a voxel size of 0.83 × 0.83 × 0.80 mm3 on a 3T MRI system to distinguish the RN, SN, and STN. A deeply supervised attention U-net was pre-trained with a dataset of 100 subjects containing susceptibility maps with a voxel size of 0.63 × 0.63 × 2.00 mm3 to provide initial weights for the target network. Five-fold cross-validation over the training cohort was used for all the models’ training and selection. The same test cohort was used for the final evaluation of all the models. Dice coefficients were used to assess spatial overlap agreement between manual delineations (ground truth) and automated segmentation. Volume and magnetic susceptibility values in the nuclei extracted with automated CNN delineation were compared to those extracted by manual tracing. Consistencies of volume and magnetic susceptibility values by different extraction strategies were assessed by Pearson correlation coefficients and Bland-Altman analyses.ResultsThe automated CNN segmentation method achieved mean Dice scores of 0.903, 0.864, and 0.777 for the RN, SN, and STN, respectively. There were no significant differences between the achieved Dice scores and the inter-rater Dice scores (p > 0.05 for each nucleus). The overall volume and magnetic susceptibility values of the nuclei extracted by the automatic CNN method were significantly correlated with those by manual delineation (p < 0.01).ConclusionMidbrain structures can be precisely segmented in high-resolution susceptibility maps using a CNN-based method.

Projected pooling loss for red nucleus segmentation with soft topology constraints

“…15-20, 51, 52 Among these, some are based on QSM data. [15][16][17][18]20 Our results are in line with those of the best performing approaches which reach around 90% of Dice accuracy. 15,17,18 However, these three studies only included healthy controls.…”

“…[15][16][17][18]20 Our results are in line with those of the best performing approaches which reach around 90% of Dice accuracy. 15,17,18 However, these three studies only included healthy controls. Thus, it is unclear how they would perform on patients.…”

“…49 The red nucleus has ferromagnetic properties and QSM is thus an imaging technique that is well-adapted to its visualization and segmentation. The dataset includes a total of 80 participants, including 18 QSM images were manually segmented by a trained neuroradiologist (L.C.). Figure 3 shows an example of QSM data with the red nucleus segmentation.…”

“…Several approaches have been proposed for automatic segmentation of the red nucleus from QSM data. [15][16][17][18][19][20] However, to the best of our knowledge, none of them has assessed the topological correctness of the results. Moreover, these papers only includes healthy participants and were thus not evaluated in patients with parkinsonism.…”

Combining simple interactivity and machine learning: a separable deep learning approach to subthalamic nucleus localization and segmentation in MRI for deep brain stimulation surgical planning

“…Although this distortion is more pronounced in cortical regions, Lau et al 10 found that the geometric distortion between 3T and 7T MRI is on the order of 1 to 2 mm around the subthalamic nucleus, rendering it more difficult to directly use 7T image registration given the lack of contrast in the 3T image to correct for these local distortions. Zhao et al 11 combined the two approaches and applied U-Nets to directly segmenting the STN on high-resolution susceptibility maps and achieved an accuracy on the order of 78% Dice, although they only used healthy subjects.…”