This study aims to develop and clinically validate a deep learning-based algorithm for predicting individual progression of geographic atrophy (GA) lesions from baseline optical coherence tomography (OCT) images in patients with age-related macular degeneration (AMD). The algorithm was evaluated using 184 eyes of 100 patients, with OCT and fundus autofluorescence (FAF) images from routine clinical care. FAF images were manually annotated by certified readers and registered to corresponding OCT scans to create 2-dimensional en face annotations. The deep learning model was trained to predict GA lesion growth over time from a single baseline OCT image. The performance metrics included Dice similarity coefficient (DSC) and mean absolute error (MAE). The model achieved a mean DSC of 0.80 for the total GA region at baseline, which increased slightly to 0.82 over the first 2 years but slightly decreased to 0.70 after 3 years. The MAE was low initially and slowly increased over time. The model also successfully identified fast progressors, with an area under the curve of 0.81, 0.79, and 0.77 for the top 10%, 15%, and 20% of GA growth rates, respectively. The study concludes that the proposed algorithm can accurately predict GA lesion growth over time, providing a valuable tool for clinical decision support and personalized treatment planning.This study aims to develop and clinically validate a deep learning-based algorithm for predicting individual progression of geographic atrophy (GA) lesions from baseline optical coherence tomography (OCT) images in patients with age-related macular degeneration (AMD). The algorithm was evaluated using 184 eyes of 100 patients, with OCT and fundus autofluorescence (FAF) images from routine clinical care. FAF images were manually annotated by certified readers and registered to corresponding OCT scans to create 2-dimensional en face annotations. The deep learning model was trained to predict GA lesion growth over time from a single baseline OCT image. The performance metrics included Dice similarity coefficient (DSC) and mean absolute error (MAE). The model achieved a mean DSC of 0.80 for the total GA region at baseline, which increased slightly to 0.82 over the first 2 years but slightly decreased to 0.70 after 3 years. The MAE was low initially and slowly increased over time. The model also successfully identified fast progressors, with an area under the curve of 0.81, 0.79, and 0.77 for the top 10%, 15%, and 20% of GA growth rates, respectively. The study concludes that the proposed algorithm can accurately predict GA lesion growth over time, providing a valuable tool for clinical decision support and personalized treatment planning.