Predictive capabilities of each model in the convolutional neural network
Model inputs | UWF colour | UWF FAF | GC-IPL | OCTA | OCT/OCTA quantitative data | Patient data | AUC on test set (95% CI) |
UWF colour | X | 0.450 (0.282 to 0.592) | |||||
UWF FAF | X | 0.618 (0.462 to 0.773) | |||||
OCTA | X | 0.582 (0.440 to 0.724) | |||||
GC-IPL | X | 0.809 (0.700 to 0.919) | |||||
Quantitative data and patient data | X | X | 0.754 (0.625 to 0.882) | ||||
All images | X | X | X | X | 0.829 (0.719 to 0.939) | ||
All images and quantitative data | X | X | X | X | X | 0.830 (0.726 to 0.940) | |
OCTA and GC-IPL | X | X | 0.828 (0.718 to 0.938) | ||||
All images and all data | X | X | X | X | X | X | 0.836 (0.729 to 0.943) |
GC-IPL, OCTA, quantitative data and patient data | X | X | X | X | 0.832 (0.724 to 0.940) | ||
GC-IPL, quantitative data and patient data* | X | X | X | 0.841 (0.739 to 0.943) |
AUC on validation and test set figures describe the performance of each model on predicting the probability of symptomatic AD diagnosis for each individual eye in the independent validation set. AUC values represent average performance over epochs 50–100. An ‘x’ indicates that the input(s) were included in the model described in that row.
*Indicates best-performing model.
AD, Alzheimer’s disease; AUC, area under the receiver operating characteristic curve; FAF, fundus autofluorescence; GC-IPL, ganglion cell-inner plexiform layer thickness maps; OCTA, optical coherence tomography angiography; UWF, ultra-widefield color images.