Abstract PO-040: Retinal disease diagnosis through an adaptive deep learning model

Abstract

Abstract This research presents an adaptive deep learning model for retinal disease classification based on 3D optical coherence tomography (OCT) images. Computer-aided retinal disorder diagnosis is one of the critical research topics that can benefit the patient care process. 3D medical images can provide more comprehensive information about retinal structures. However, due to the higher data dimension and more complicated data pattern compared to 2D images, 3D image analysis is challenging the performance of many machine learning models. This research aims to develop a machine learning model for retinal disease diagnosis on 3D OCT images. A public dataset, which includes age-related macular degeneration (AMD), diabetic macular edema (DME), and the normal cases, is used in this study. The model includes three main procedures: 1) Image preprocessing is conducted for each cross-sectional retinal image. A novel Hough transform-based line detection approach is developed to extract 3D retinal layers. 2) 3D retinal layer samples are constructed, which are used to optimize an adaptive deep learning model to identify the most effective and efficient model structure. A novel objective function is proposed to optimize the convergence trend in this process. The constructed deep learning structure is based on the search result of a neural architecture space. 3) The optimized deep learning model is trained using the extracted 3D samples. The experimental results show that the proposed model can significantly boost the training efficiency and improve the testing accuracy. Citation Format: Haifeng Wang. Retinal disease diagnosis through an adaptive deep learning model [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-040.