gms | German Medical Science

Objective: Brain metastases (BM) are the most common brain tumours. During the resection of BM, it is essential to precisely distinguish between tumour and surrounding brain tissue (BT) as tiny tumour remnants might lead to local recurrence. Optical coherence tomography (OCT) is an imaging technique with a resolution of a few micrometres that might improve the detection of small tumour remnants. The objective of this study was to distinguish the necrotic and vital proportion of BM from healthy BT in an ex vivo setting by applying texture analysis and machine learning algorithms for tissue classification to OCT images.

Methods: Twenty-eigtht patients with suspected BM were included in the study. BM tissue samples were taken during resection using neuro-navigation and microsurgical techniques. BT was taken during the approach in subcortical lesions only. Fresh tissue samples were scanned using a spectral domain high-resolution OCT system. Volumetric OCT images were stored for further post-processing. Tissue samples were then evaluated histologically. The percentage of vital tumour tissue, necrosis and BT per sample was recorded. Texture feature-based post-processing as well as machine learning with principal component analysis (PCA) and support vector machines (SVM) were applied to the OCT scans. Ultimately, a classification using SVM was carried out, determining the accuracy with which BM tissue (sample with at least 60% of vital tumour, alternatively samples with 100% necrotic tissue) can be distinguished from BT (at least 90% healthy tissue).

Results: Every possible combination of used texture parameters according to the PCA and SVM was considered. 6 BM samples showed 100% of necrotic tissue. Those were compared to BT. The SVM enabled us to classify necrotic and healthy tissue with an accuracy of 99.07%. The best result was obtained for a combination of local binary pattern and Law’s energy texture measure. The same procedure was applied for those samples of at least 60% vital tumour tissue (n=7). It was possible to classify vital tumour against healthy tissue with an accuracy of 95.75%.

Conclusion: This innovative method of OCT and texture-based classification enables us to precisely distinguish between healthy tissue on the one hand and tumorous necrotic and vital tissue on the other hand. This is a promising method to be tested in an intraoperative setting.