LoAd: a locally adaptive cortical segmentation algorithm

Abstract

Thickness measurements of the cerebral cortex can aid diagnosis and provide valuable information about the temporal evolution of diseases such as Alzheimer's, Huntington's, and schizophrenia. Methods that measure the thickness of the cerebral cortex from in-vivo magnetic resonance (MR) images rely on an accurate segmentation of the MR data. However, segmenting the cortex in a robust and accurate way still poses a challenge due to the presence of noise, intensity non-uniformity, partial volume effects, the limited resolution of MRI and the highly convoluted shape of the cortical folds. Beginning with a well-established probabilistic segmentation model with anatomical tissue priors, we propose three post-processing refinements: a novel modification of the prior information to reduce segmentation bias; introduction of explicit partial volume classes; and a locally varying MRF-based model for enhancement of sulci and gyri. Experiments performed on a new digital phantom, on BrainWeb data and on data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) show statistically significant improvements in Dice scores and PV estimation (p<10(-3)) and also increased thickness estimation accuracy when compared to three well established techniques.

Fig. 1

Segmentation of a BrainWeb T1-weighted dataset with 3% noise and 20% INU: left) BrainWeb ground truth segmentation; centre) MAP with MRF but without the proposed improvements; right) proposed method.

Fig. 2

MRF class connectivity network.

Fig. 3

Algorithm flowchart.

Fig. 4

The mixed class prior (dashed green) is the normalised geometric mean of p_ik and p_ij (dashed blue and red respectively). The continuous lines represent their value after normalisation over all classes.

Fig. 5

Sulci localisation using the proposed metric. (a) Current binary segmentation, (b) hard segmented set in green with the respective speed function s_j in grey levels, (c) geodesic distance (time of arrival), (d) the phantom in red overlaid with the detected sulci location in white.

Fig. 6

Sulci and gyri enhancement: (left) expected segmentation; (centre) (h_CSF, s_WM) and (h_WM, s_CSF) on the top and bottom respectively; (right) ${\omega }_i^{\text{sulci}}$ and $a_i^{\text{gyri}}$ in green and red respectively.

Fig. 7

(Left) The MNI305 atlas and (right) the ICBM452.

Fig. 8

(Top) The fuzzy Dice scores between the cortical GM segmentations using different atlas and relaxation factors. Segmentation example with RelaxationFactor = 0 (bottom left) and Relaxation Factor = 1 (bottom right). Notice the improved segmentation results in the ventricle area.

Fig. 9

Phantom segmentation and thickness results: a) 3D model of the phantom, b) high noise phantom, c) true labels and GM prior used, d) ML without MRF, e) ML with MRF, f) proposed method. The red arrows point to the presence of noise and lack of detail causing wrong thickness measurements. The green arrows point to the detected deep gyri.

Fig. 10

(a) Normalised cumulative histogram of the absolute difference between the segmentation and the ground truth; (b) Dice score between the segmentation and the ground truth at several threshold values.

Fig. 11

Statistical significance of cortical thickness between AD patients and controls: colour coded p-values are represented in logarithmic scale with positive and negative values associated with thinning and thickening respectively.