Polarimetric biomarkers of peri-tumoral stroma can correlate with 5-year survival in patients with left-sided colorectal cancer

Abstract

Using a novel variant of polarized light microscopy for high-contrast imaging and quantification of unstained histology slides, the current study assesses the prognostic potential of peri-tumoral collagenous stroma architecture in 32 human stage III colorectal cancer (CRC) patient samples. We analyze three distinct polarimetrically-derived images and their associated texture features, explore different unsupervised clustering algorithm models to group the data, and compare the resultant groupings with patient survival. The results demonstrate an appreciable total accuracy of ~ 78% with significant separation (p < 0.05) across all approaches for the binary classification of 5-year patient survival outcomes. Surviving patients preferentially belonged to Cluster 1 irrespective of model approach, suggesting similar stromal microstructural characteristics in this sub-population. The results suggest that polarimetrically-derived stromal biomarkers may possess prognostic value that could improve clinical management/treatment stratification in CRC patients.

Figure 1

Whole-slide and ROI-based histologic and polarimetric imaging of a stage III CRC sample for qualitative ROI-based analysis. (a) H&E-stained slide and (b) corresponding polarimetric intensity image of an adjacent unstained slide (see Polarimetric image quantification & analysis section). (c) H&E and (d) polarimetric intensity zoomed-in images of the region around the right-most ROI in (a) and (b). The brighter areas visualize birefringent tissues that contain more collagenous stroma. The green squares represent the ROIs identified by the pathologist on the H&E images at the leading edges of the tumour, and the arrow show their transfer onto the polarimetric images. Images were obtained at 20 × magnification and were tiled and stitched for (a) and (b), and at 80 × for (c) and (d).

Figure 2

Workflow for proposed polarimetric imaging, analysis, and unsupervised clustering pipeline. (a) widefield polarimetric imaging of unstained tissue sample acquired at 18 angular orientations of the crossed linear polarizers (5° increments); (b) selection and calculation of novel polarimetric stromal features on the derived Intensity, Alignment and Density images from the 200 × 200 μm ROI identified by pathologist (green square); (c) calculation of median polarimetric values and GLCM texture features (contrast, correlation, energy, homogeneity and entropy) within ROI (red square), applied to each of the three derived polarimetric images; (d) two-step normalization to ensure feature values are zero-centered and resemble a normal distribution, as needed for clustering model input; (e) clustering of the ROIs using the polarimetric and texture features, with the resultant outputs clusters then assessed for correlations with patient outcomes.

Figure 3

Resultant cluster assignments (blue = Cluster 1, orange = Cluster 2) of each region (n = 297) across all 32 patients for (a) K-means model, (b) Fuzzy C-means, (c) Gaussian Mixture Model. The black line partitions those who survived from those that did not. The number above each bar indicates the overall cluster assignment for that patient based on the majority-vote approach; the ambiguous “perfect tie” patients (*) remained unassigned (for details, see text). The table summarizes the assignment data for all 297 ROIs.

Figure 3

Resultant cluster assignments (blue = Cluster 1, orange = Cluster 2) of each region (n = 297) across all 32 patients for (a) K-means model, (b) Fuzzy C-means, (c) Gaussian Mixture Model. The black line partitions those who survived from those that did not. The number above each bar indicates the overall cluster assignment for that patient based on the majority-vote approach; the ambiguous “perfect tie” patients (*) remained unassigned (for details, see text). The table summarizes the assignment data for all 297 ROIs.

Figure 4

Cluster-Patient groupings (n = 32) using the majority-vote approach as determined by the three unsupervised clustering model. (a) K-means; (b) Fuzzy C-means; (c) Gaussian Mixture model. Green represents patients which survived after 5-years (n = 23) whereas purple denotes patients which did not (n = 9). Of note is the composition of Cluster 1 across all models, being primarily comprised of 5-year survivors (for details, see text).

Figure 5

Cluster-Patient groupings (n = 32) using the features-averaged approach as determined by the three unsupervised clustering model. (a) K-means; (b) Fuzzy C-means; (c) Gaussian Mixture model. Green represents patients which survived after 5-years (n = 23) whereas purple denotes patients which did not (n = 9). Of note is the composition of Cluster 1 across all models, being primarily comprised of 5-year survivors (for details, see text).

Figure 6

Two ROIs (green squares) on polarimetric intensity images of patient #30. (a) was deemed to belong to Cluster 1 with a weight of 90%, in comparison to (b) which was also considered to belong to Cluster 1, but with a weight of 56% (Fuzzy C-means algorithm). There are visible structural differences as is indeed reflected in the derived weights; this relative weight information could be used in future model refinements.