Tuning of optical coherence tomography texture features as a basis for tissue differentiation in glioblastoma samples

In this work we configure Haralick's texture parameter contrast to match the dimensions of characteristic structural features in ex vivo samples from glioblastoma (GBM) resection imaged with optical coherence tomography (OCT). The aim is to find and tune different texture features in a way that they enable the best possible basis for tissue classification using support vector machines (SVM). We used a sample collective including 18 tissue samples comprising 9 samples with at least 90% vital tumor and no healthy tissue, as well as 9 samples with 100% healthy tissue. All samples were imaged ex vivo immediately after resection. As a reference all samples were then examined professionally in the department of histopathology to determine tissue percentages. Based on the acquired 3D OCT images, texture features were extracted and optimized supported by the knowledge of medical professionals. Relations between the size of characteristic structures in healthy tissue as well as in GBM and different texture features were examined and evaluated. We focused on texture parameters as proposed by Haralick, relying on gray-level co-occurrence matrices (GLCMs). The displacement vector for the determination of those GLCMs was matched with size and direction of the characteristic structural tissue features of healthy and tumorous tissue. The results serve as a starting point to optimize the classification process of GBM against healthy tissue using SVM.

Automated summary

This work utilizes optical coherence tomography to image glioblastoma resection specimens and optimizes texture features for tissue classification. The study highlights the relationships between texture features and structural characteristics in healthy and tumor tissues.