This AI Paper Introduces a Deep Learning Model for Classifying Stages of Age-Related Macular Degeneration Using Real-World Retinal OCT Scans

A new research paper presents a deep learning-based classifier for age-related macular degeneration (AMD) stages using retinal optical coherence tomography (OCT) scans. Utilizing a two-stage convolutional neural network, the model classifies macula-centered 3D volumes from Topcon OCT images into Normal, early/intermediate AMD (iAMD), atrophic (GA), and neovascular (nAMD) stages. The first stage employs a 2D ResNet50 for B-scan classification, and the second stage uses smaller models (ResNets) for volume classification. 

The model, trained on a substantial dataset, performs strongly in categorizing macula-centered 3D volumes into Normal, iAMD, GA, and nAMD stages. The study emphasizes the significance of accurate AMD staging for timely treatment initiation. Performance metrics include ROC-AUC, balanced accuracy, accuracy, F1-Score, sensitivity, specificity, and Matthews correlation coefficient.

The research details creating a deep learning-based system for automated AMD detection and staging using retinal OCT scans. OCT, a non-invasive imaging technique, is crucial in providing detailed insights into AMD staging compared to traditional methods. The study emphasizes the significance of accurate AMD staging for effective treatment and vision preservation. The research highlights the importance of high-quality datasets for robust analysis.

The study implemented a two-stage deep learning model utilizing ImageNet-pretrained ResNet50 and four separate ResNets for binary classification of AMD biomarkers on OCT scans. The first stage localized disease categories within the volume, while the second stage performed volume-level classification. The model, trained on a real-world OCT dataset, demonstrated promising performance metrics, including ROC-AUC, balanced accuracy, accuracy, F1-Score, sensitivity, specificity, and Matthews correlation coefficient. The study acknowledged challenges in using diverse OCT datasets from different devices and discussed potential generalization issues.

The deep learning-based AMD detection and staging system demonstrated promising performance with an average ROC-AUC of 0.94 in a real-world test set. Incorporating Monte-Carlo dropout at inference time enhanced the reliability of classification uncertainty estimates. The study utilized a curated dataset of 3995 OCT volumes from 2079 eyes, evaluating performance with various metrics, including AUC, BACC, ACC, F1-Score, sensitivity, specificity, and MCC. The results highlight the model’s potential for accurate AMD classification and staging, aiding in timely treatment and visual function preservation.

The study successfully developed an automated deep learning-based AMD detection and staging system using OCT scans. The two-stage convolutional neural network accurately classified macula-centered 3D volumes into four classes: Normal, iAMD, GA, and nAMD. The deep learning model showed comparable or better performance than baseline approaches, with the additional benefit of B-scan-level disease localization.

Further research can enhance the deep learning model’s generalizability to various OCT devices, considering adaptations for scanners like Cirrus and Spectralis. Domain shift adaptation methods should be explored to address limitations related to dataset-specific training, ensuring robust performance across diverse signal-to-noise ratios. The model’s potential for retrospective AMD onset detection could be extended, allowing automatic labeling of longitudinal datasets. Application of uncertainty estimates in real-world screening settings and exploring the model for detecting other disease biomarkers beyond AMD are promising avenues for future investigation, aiding disease screening in a broader population.

Check out the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join our 33k+ ML SubReddit, 41k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..