The study aimed to create a machine learning method for differentiating diagnostically valued tumorous tissue from diagnostically “non-valued” non-tumorous tissues in the human brain, using cross-polarization optical coherence tomography (CP OCT) in order to provide guidance for stereotactic biopsies. A method of feature extraction from OCT data in two orthogonal polarization channels has been proposed and a classification algorithm for the resulting feature vectors has been created. If used for stereotactic biopsy guidance, the proposed approach could decrease the number of excised diagnostically non-valued samples and minimize the invasiveness of the procedure and the risk of excessive bleeding.
Development of a literature informed Bayesian machine learning method for feature extraction and classification