Texture analysis of 18F-florbetapir PET brain images for the diagnosis of Alzheimer’s disease

Resumen

Alzheimer’s disease is characterised by pathological plaques outside the neurons formed by amyloid-beta(Aβ)that start occurring in the preclinical phase of the disease. PET imaging based on Aβ-binding radiotracers is used in the diagnosis of AD. These include 11C-Pittsburgh compound B and fluorine-labelled tracers like florbetapir (FBP).The images are visually analysed and classified into amyloid negative(A-) and amyloid positive(A+). This classification is based on the uptake of the radiotracer in cortical brain regions and the difference to the adjacent white matter.Quantitative feature extraction of amy-loid PET images is proposed to help in the classification of difficult cases. First, the images are segmented into cortical brain regions. Then, Standard Uptake Value ratios(SUVR) and textural features based on the grey level co-occurrence matrix (GLCM) are extracted from the images. AnSVM model is computed to classify amyloid PET images based on the extracted features. SUVRs, textural features and a combination of both are evaluated. The results show thatfeature vectors composed of 9 textural features offer the highest prediction accuracy, sensitivity and specificity(0.97, 0.94 and 1.00, respectively). Therefore, textural features are shown to be potential image features to correctly classify PET-amyloid images into A-and A+.