Syntactic structure analysis in uveal melanomas

Abstract

Syntactic structure analysis was carried out successfully on 92 paraffin embedded uveal melanomas, taken from patients with a minimum follow up of 5 years. This simple, fast, and reproducible method of describing the tumour architecture has been significantly correlated with malignancy in tumours from several sites. Paraffin sections 5 microns thick, were cut and stained with haematoxylin and eosin. Tumours were classified according to a modification of the Callender classification. A minimum spanning tree (MST), using the centre points of tumour nuclei, was constructed in five randomly chosen fields with an interactive digitising video overlay system. Ten syntactic structure features were derived from each MST; subsequently, the mean and standard deviation of the five fields analysed were calculated for further statistical analysis. Reproducibility was acceptable with a mean correlation coefficient of 0.70. In univariate survival analysis, the percentage of points with three neighbours yielded prognostic significance (p < 0.05). Minimum spanning tree variables were compared (chi 2 test) with classic tumour prognosticators and there was a significant correlation between Callender cell type and the following MST parameters: mean number of points (p < 0.003); MST length (p < 0.003); mean line length (p < 0.01); number of nuclei with one neighbour (p < 0.004); number of nuclei with two neighbours (p < 0.02), and number of nuclei with three neighbours (p < 0.005). Syntactic structure analysis is an evolving technique, but may be able to mathematically (and reproducibly) describe melanoma architecture across the spectrum of the Callender classes. This would also allow architectural grading of tumours within the specific Callender groups, providing more precise prognostic information. Further modifications of this technique are necessary to optimise prognostic potential when applied to uveal melanomas.