Add to Chrome
Add to Firefox
Add to EdgeA Gabor Filter Texture Analysis Approach for Histopathological Brain Tumor Subtype Discrimination
Apr 17, 2017
Meningioma brain tumour discrimination is challenging as many histological patterns are mixed between the different subtypes. In clinical practice, dominant patterns are investigated for signs of specific meningioma pathology; however the simple observation could result in inter- and intra-observer variation due to the complexity of the histopathological patterns. Also employing a computerised feature extraction approach applied at a single resolution scale might not suffice in accurately delineating the mixture of histopathological patterns. In this work we propose a novel multiresolution feature extraction approach for characterising the textural properties of the different pathological patterns (i.e. mainly cell nuclei shape, orientation and spatial arrangement within the cytoplasm). The pattern textural properties are characterised at various scales and orientations for an improved separability between the different extracted features. The Gabor filter energy output of each magnitude response was combined with four other fixed-resolution texture signatures (2 model-based and 2 statistical-based) with and without cell nuclei segmentation. The highest classification accuracy of 95% was reported when combining the Gabor filters energy and the meningioma subimage fractal signature as a feature vector without performing any prior cell nuceli segmentation. This indicates that characterising the cell-nuclei self-similarity properties via Gabor filters can assists in achieving an improved meningioma subtype classification, which can assist in overcoming variations in reported diagnosis.
