Deep-Learning-gestützte Computer-Algorithmen zur Diagnosestellung in der Augenheilkunde

Abstract

<b><i>Purpose:</i></b> The advent of electronic medical records (EMRs) with large electronic imaging databases along with advances in deep neural networks with machine learning has provided a unique opportunity to achieve milestones in automated image analysis. Optical coherence tomography is the most common imaging modality in ophthalmology and represents a dense and rich data set when combined with labels derived from the EMR. We sought to determine whether deep learning could be utilized to distinguish normal OCT images from images from patients with age-related macular degeneration (AMD). <b><i>Design:</i></b> EMR and OCT database study. <b><i>Subjects:</i></b> Normal and AMD patients who underwent macular OCT. <b><i>Methods:</i></b> Automated extraction of an OCT database was performed and linked to clinical end points from the EMR. Optical coherence tomography scans of the macula were obtained by Heidelberg Spectralis, and each OCT scan was linked to EMR clinical end points extracted from EPIC. The central 11 images were selected from each OCT scan of 2 cohorts of patients: normal and AMD. Cross-validation was performed using a random subset of patients. Receiver operating characteristic (ROC) curves were constructed at an independent image level, macular OCT level, and patient level. <b><i>Main Outcome Measure:</i></b> Area under the ROC curve. <b><i>Results:</i></b> Of a recent extraction of 2.6 million OCT images linked to clinical data points from the EMR, 52 690 normal macular OCT images and 48 312 AMD macular OCT images were selected. A deep neural network was trained to categorize images as either normal or AMD. At the image level, we achieved an area under the ROC curve of 92.78% with an accuracy of 87.63%. At the macula level, we achieved an area under the ROC curve of 93.83% with an accuracy of 88.98%. At a patient level, we achieved an area under the ROC curve of 97.45% with an accuracy of 93.45%. Peak sensitivity and specificity with optimal cutoffs were 92.64% and 93.69%, respectively. <b><i>Conclusions:</i></b> The deep learning technique achieves high accuracy and is effective as a new image classification technique. These findings have important implications in utilizing OCT in automated screening and the development of computer-aided diagnosis tools in the future.