Answers are grounded in this paper and cite their sources.
Purpose To gain an understanding of data labeling requirements to train deep learning models for measurement of geographic atrophy with autofluorescence (FAF) images. Design Evaluation of AI algorithms Subjects The Age-Related Eye Disease Study 2 images were used for training and cross validation and GA clinical trial images were used for testing. Methods Training data consisted of two sets of FAF images; one with area measurements only and no indication of GA location (Weak-labeled) and the second with GA segmentation masks (Strong-labeled). Main outcome measures Bland Altman plots and scatter plots were used to compare GA area measurement between ground truth and AI measurements. Dice Coefficient was used to compare accuracy of segmentation of Strong model. Results In the cross validation AREDS2 dataset (n = 601), mean area (SD) of GA measured by human grader, Weak-labeled and Strong-labeled AI models was 6.65 mm 2 (6.3), 6.83 mm 2 (6.29) and 6.58 mm 2 (6.24). The mean difference between ground truth and AI was 0.18 mm 2 (95% CI -7.57,7.92) for the Weak-labeled model and -0.07 mm 2 (-1.61,1.47) for the Strong-labeled model. With GSK testing data (n = 156), the GA area (SD) was 9.79 mm 2 (SD 5.6), 8.82 mm 2 (4.61) and 9.55 mm 2 (5.66) respectively. The mean difference between ground truth and AI for the 2 models was -0.97 mm 2 (- 4.36,2.41) and -0.24 mm 2 (- 4.98,4.49) respectively. The Dice Coefficient was 0.99 for intergrader agreement, 0.88 for the cross-validation data and 0.92 for the testing data. Conclusions Deep learning models can achieve reasonable accuracy even with Weakly-labeled data. Training methods that integrate large volumes of Weak-labeled images with small number of Strong-labeled images offer a promising solution to overcome the burden of cost and time for data labeling. Words 3290
No takes yet. Share an insight, caveat, or question.
Domalpally et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: