End-to-End Deep Learning Model for Predicting Treatment Requirements in Neovascular AMD From Longitudinal Retinal OCT...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Neovascular age-related macular degeneration (nAMD) treatment with anti-VEGF drugs shows variable patient responses.
• Current treatment frequency for nAMD is difficult to predict, leading to suboptimal outcomes.
• Optical coherence tomography (OCT) provides crucial imaging data for nAMD management.

Purpose of the Study:

• To develop a predictive model for anti-VEGF treatment requirements in nAMD patients.
• To utilize automated image analysis of longitudinal OCT scans for personalized treatment prediction.
• To improve treatment planning and frequency for nAMD patients.

Main Methods:

• A deep learning (DL) model combining DenseNet and RNN architectures was developed.
• The model processes 2D image slices from OCT volumes to extract spatial and temporal features.
• The DL model was trained on 281 nAMD patients and validated on 69 patients using a pro-re-nata regimen.

Main Results:

• The DL model achieved a concordance index of 0.7 for predicting treatment numbers.
• It demonstrated an AUC of 0.85 for detecting low treatment requirement patients and 0.81 for high requirement patients.
• The proposed DL approach outperformed previous machine learning methods relying on handcrafted features.

Conclusions:

• Deep learning models can effectively extract spatio-temporal patterns from raw OCT images for nAMD treatment prediction.
• Automated analysis of OCT scans holds significant potential for optimizing anti-VEGF therapy in nAMD.
• This predictive model offers a pathway towards more personalized and efficient nAMD management.