Three-dimensional volume reconstruction of extracellular matrix proteins in uveal melanoma from fluorescent confocal laser scanning microscope images

Abstract

The distribution of looping patterns of laminin in uveal melanomas and other tumours has been associated with adverse outcome. Moreover, these patterns are generated by highly invasive tumour cells through the process of vasculogenic mimicry and are not therefore blood vessels. Nevertheless, these extravascular matrix patterns conduct plasma. The three-dimensional (3D) configuration of these laminin-rich patterns compared with blood vessels has been the subject of speculation and intensive investigation. We have developed a method for the 3D reconstruction of volume for these extravascular matrix proteins from serial paraffin sections cut at 4 microm thicknesses and stained with a fluorescently labelled antibody to laminin (Maniotis et al., 2002). Each section was examined via confocal laser-scanning focal microscopy (CLSM) and 13 images were recorded in the Z-dimension for each slide. The input CLSM imagery is composed of a set of 3D sub-volumes (stacks of 2D images) acquired at multiple confocal depths, from a sequence of consecutive slides. Steps for automated reconstruction included (1) unsupervised methods for selecting an image frame from a sub-volume based on entropy and contrast criteria, (2) a fully automated registration technique for image alignment and (3) an improved histogram equalization method that compensates for spatially varying image intensities in CLSM imagery due to photo-bleaching. We compared image alignment accuracy of a fully automated method with registration accuracy achieved by human subjects using a manual method. Automated 3D volume reconstruction was found to provide significant improvement in accuracy, consistency of results and performance time for CLSM images acquired from serial paraffin sections.

Fig. 1

An overview of 3D volume reconstruction steps from input fluorescent confocal laser scanning microscope 3D subvolumes. The processing steps start with a set of subvolumes and end with one registered 3D volume.

Fig. 2

Illustration of disc-based segmentation. A segment is formed as a connected set of pixels covered by a disc while the disk moves within the fluorescent boundary. Depending on the disc diameter and contour gaps a segment is either detected or not.

Fig. 3

Illustration of multiple contour interpretations. A partially closed contour could lead to three detection outcomes depending on the disc diameter.

Fig. 4

Illustrations of the case when distance-based matching leads to an erroneous match of the segments labelled as 3 (left) and as 4 (right).

Fig. 5

Illustration of the correspondence problem for two sets of features S_i and T_j with unequal number of features N^S = 3 (left) and N^T = 5 (right). Segments are shown as discs characterized by their index i, area $a_i^S$ and centroid location ${\overrightarrow{c}}_i^S$. Dashed lines represent the Euclidean distances between any two centroid locations.

Fig. 6

Illustration of distance-based matching for two pivot segments $S_i=\{{\overrightarrow{c}}_i^S,a_i^S\}$ and $T_j=\{{\overrightarrow{c}}_j^T,a_j^T\}$. In order to find the best match, distance and area ratio of pairs of segments are compared to satisfy D_ij(u, v) < δ₁ and A_ij(u, v) < ε₁.

Fig. 7

Real and virtual subvolumes of a specimen show two possible cuts of a specimen.

Fig. 8

Illustration of the three selected pairs of evaluation images.

Fig. 9

Evaluation of image frame selection from within each 3D subvolume to be used for alignment of the subvolumes. The graphs show combined visual saliency score as a function of image frame for two 3D volumes. The frame with maximum visual saliency score within each subvolume would be used for alignment.

Fig. 10

Evaluation of interframe similarity for two spatially adjacent subvolumes with 22 frames after alignment. The similarity is measured by normalized correlation (vertical axis). The horizontal axis refers to the pairs of frames that start with the end frames of subvolumes (21–0 ∼ – ) and finish with the middle frames (11–10 ∼ – ).

Fig. 11

Alignment error as a function of compactness measure associated with each trial. A high compactness measure implies that the control points selected during manual alignment are spatially dense or close to being collinear, thus leading to large alignment error. A hypothetical linear relationship of the variables is shown.

Fig. 12

Two representative images that have to be aligned. The mathematical notations of these images in the text is $I_a^S\in V^S$ (left) and $I_b^T\in V^T$ (right). Human tonsil tissue stained for laminin.

Fig. 13

Segmentation of input images shown in Fig. 12. Segmentation is performed by thresholding followed by connectivity analysis with a disc. The two images illustrate results obtained with different disc parameters [left image – T.S. (threshold S) = 10, right image – T.T. (threshold T) = 8, M.R. (minimum size of a region) = 80, and D.D. (disc diameter) = 1].

Fig. 14

The correspondence outcome from two phases for the segments shown in Fig. 13. The left and right images are to be aligned. Overlays illustrate established correspondences between segments that are labelled from 1 to 17. The centroid locations of segments are sorted based on the correspondence error from the smallest to the largest.

Fig. 15

The result of automated feature selection for the original images shown in Fig. 13 after they were processed to establish segment correspondences shown in Fig. 14. The different coloured circles represent three pairs of centroids selected automatically according to the compactness measure defined in Eq. (7).

Fig. 16

Comparison of two image enhancement techniques. Original CLSM image (left) and the results obtained by histogram equalization (middle), and by the proposed improved histogram equalization method (right) with background subtraction (threshold valuw ω = 20).