Abstract / Summary
To study a set of textural features based on optical coherence tomography (OCT) images using texture analysis via machine learning to demonstrate their potential for early identification of macular degeneration in high myopia (HM). Two hundred and eighty-eight eyes of 166 participants were included in this study as internal dataset, which was used to identify early visual impairment and myopic maculopathy in patients with HM. An external validation dataset which contained 40 eyes of 20 participants was used to test the performance of the classification model. Texture analysis was performed to extract textural features of OCT images. The area under the receiver operating characteristic curve (AUC), accuracy, precision and recall were used to evaluate the classification performance of the features. The Fisher’s score was used to rank the discriminative power of each feature. Correlations between textural features and the corresponding thicknesses were determined. To better understand the algorithms, visualization heatmap analysis was performed. Approximately 144 textural features were discovered in macular OCT images, and the performance of textural features combined with thickness was better than thickness only in the classification system. And the inclusion of demographic characteristics could improve discriminative power of the classification system further. A few textural features of the outer retina were independent of the corresponding thicknesses. Textural features can be helpful to identify macular degeneration earlier and more effectively by combing with traditional clinical features in patients with HM.