Addressing Artificial Intelligence Bias in Retinal Diagnostics

Abstract

Purpose This study evaluated generative methods to potentially mitigate artificial intelligence (AI) bias when diagnosing diabetic retinopathy (DR) resulting from training data imbalance or domain generalization, which occurs when deep learning systems (DLSs) face concepts at test/inference time they were not initially trained on. Methods The public domain Kaggle EyePACS dataset (88,692 fundi and 44,346 individuals, originally diverse for ethnicity) was modified by adding clinician-annotated labels and constructing an artificial scenario of data imbalance and domain generalization by disallowing training (but not testing) exemplars for images of retinas with DR warranting referral (DR-referable) from darker-skin individuals, who presumably have greater concentration of melanin within uveal melanocytes, on average, contributing to retinal image pigmentation. A traditional/baseline diagnostic DLS was compared against new DLSs that would use training data augmented via generative models for debiasing. Results Accuracy (95% confidence intervals [CIs]) of the baseline diagnostics DLS for fundus images of lighter-skin individuals was 73.0% (66.9% to 79.2%) versus darker-skin of 60.5% (53.5% to 67.3%), demonstrating bias/disparity (delta = 12.5%; Welch t -test t = 2.670, P = 0.008) in AI performance across protected subpopulations. Using novel generative methods for addressing missing subpopulation training data (DR-referable darker-skin) achieved instead accuracy, for lighter-skin, of 72.0% (65.8% to 78.2%), and for darker-skin, of 71.5% (65.2% to 77.8%), demonstrating closer parity (delta = 0.5%) in accuracy across subpopulations (Welch t -test t = 0.111, P = 0.912). Conclusions Findings illustrate how data imbalance and domain generalization can lead to disparity of accuracy across subpopulations, and show that novel generative methods of synthetic fundus images may play a role for debiasing AI. Translational Relevance New AI methods have possible applications to address potential AI bias in DR diagnostics from fundus pigmentation, and potentially other ophthalmic DLSs too.

Deepfakes in Ophthalmology

“…We found 20 published implementations of GANs specific to ophthalmology. Of these, 11 manuscripts synthesized fundus images, 9,19,20,23,32,37,[40][41][42][43][44] 6 manuscripts synthesized OCT images, [27][28][29][45][46][47] 2 manuscripts synthesized fluorescein angiography images, 48,49 and 1 manuscript synthesized infrared images 21 (Table 2). The majority of GANs were proof-ofconcept studies demonstrating feasibility of generating realistic-appearing synthetic images, specific implementations of GANs were published in 9 for diagnosis of ophthalmic diseases, including diabetic retinopathy (DR), 9,20,32,40 glaucoma, 28,45 age-related macular degeneration (AMD), 19,46 and meibomian gland dysfunction.…”

“…Moreover, it is essential to train on diverse datasets with heterogeneous features present in real-world populations to avoid biased performance in practice. 8,9 Development of these datasets typically requires sharing data across institutions, which can be limited by time, cost, legislation, 10 and privacy regulations. 11 Data-and model-sharing methods including federated 12,13 and distributed 14,15 learning have shown potential in facilitating DL algorithm training without inter-institutional data sharing.…”

“…16 Deepfakes have garnered notoriety in the media for their nefarious applications, 17,18 but recently have been explored in multiple medical domains. 9,[19][20][21][22][23][24][25][26] Since ophthalmology has been at the forefront of the DL revolution, there are numerous potential applications of synthetic images, starting with fundus 9,19,20 and optical coherence tomography (OCT). [27][28][29] Synthetic images can be modified to adjust image features such as pigmentation, 9 image quality, 30 and even disease severity.…”

“…9,[19][20][21][22][23][24][25][26] Since ophthalmology has been at the forefront of the DL revolution, there are numerous potential applications of synthetic images, starting with fundus 9,19,20 and optical coherence tomography (OCT). [27][28][29] Synthetic images can be modified to adjust image features such as pigmentation, 9 image quality, 30 and even disease severity. 31 One of many potential applications is as an alternative solution to increase the size and diversity of training datasets for DL algorithms.…”

Least kth-Order and Rényi Generative Adversarial Networks

“…Compared to other approaches, GANs have generated the most interest (e.g., see surveys in Creswell et al, 2018;Wang, She, & Ward, 2020;Wiatrak, Albrecht, & Nystrom, 2020). Generative models are used in reinforcement learning, time series predictions, fairness and privacy in artificial intelligence (AI) (Burlina, Joshi, Paul, Pacheco, & Bressler, 2020), and disentanglement (Paul, Wang, Alajaji, & Burlina, 2021), and can also be trained in a semisupervised manner, where labels and training examples are missing. Furthermore, these models are designed to produce several different outputs that are equally acceptable (Goodfellow, 2016;Karras, Laine, & Aila, 2019).…”