6 October 2025 in Nature Medicine
14 June 2025 in npj Digital Medicine
The ethical integration of artificial intelligence (AI) in healthcare necessitates addressing fairness. AI fairness involves mitigating biases in AI and leveraging AI to promote equity. Despite advancements, significant disconnects persist between technical solutions and clinical applications. Through evidence gap analysis, this review systematically pinpoints the gaps at the intersection of healthcare contexts—including medical fields, healthcare datasets, and bias-relevant attributes (e.g., gender/sex)—and AI fairness techniques for bias detection, evaluation, and mitigation. We highlight the scarcity of AI fairness research in medical domains, the narrow focus on bias-relevant attributes, the dominance of group fairness centering on model performance equality, and the limited integration of clinician-in-the-loop to improve AI fairness. To bridge the gaps, we propose actionable strategies for future research to accelerate the development of AI fairness in healthcare, ultimately advancing equitable healthcare delivery.
9 July 2021 in British Journal of Ophthalmology
Background/aims To evaluate the performances of deep learning (DL) algorithms for detection of presence and extent pterygium, based on colour anterior segment photographs (ASPs) taken from slit-lamp and hand-held cameras. Methods Referable pterygium was defined as having extension towards the cornea from the limbus of >2.50 mm or base width at the limbus of >5.00 mm. 2503 images from the Singapore Epidemiology of Eye Diseases (SEED) study were used as the development set. Algorithms were validated on an internal set from the SEED cohort (629 images (55.3% pterygium, 8.4% referable pterygium)), and tested on two external clinic-based sets (set 1 with 2610 images (2.8% pterygium, 0.7% referable pterygium, from slit-lamp ASP); and set 2 with 3701 images, 2.5% pterygium, 0.9% referable pterygium, from hand-held ASP). Results The algorithm’s area under the receiver operating characteristic curve (AUROC) for detection of any pterygium was 99.5%(sensitivity=98.6%; specificity=99.0%) in internal test set, 99.1% (sensitivity=95.9%, specificity=98.5%) in external test set 1 and 99.7% (sensitivity=100.0%; specificity=88.3%) in external test set 2. For referable pterygium, the algorithm’s AUROC was 98.5% (sensitivity=94.0%; specificity=95.3%) in internal test set, 99.7% (sensitivity=87.2%; specificity=99.4%) in external set 1 and 99.0% (sensitivity=94.3%; specificity=98.0%) in external set 2. Conclusion DL algorithms based on ASPs can detect presence of and referable-level pterygium with optimal sensitivity and specificity. These algorithms, particularly if used with a handheld camera, may potentially be used as a simple screening tool for detection of referable pterygium. Further validation in community setting is warranted. Synopsis/precis DL algorithms based on ASPs can detect presence of and referable-level pterygium optimally, and may be used as a simple screening tool for the detection of referable pterygium in community screenings.
9 September 2020 in British Journal of Ophthalmology
Background The ability of deep learning (DL) algorithms to identify eyes with neovascular age-related macular degeneration (nAMD) from optical coherence tomography (OCT) scans has been previously established. We herewith evaluate the ability of a DL model, showing excellent performance on a Korean data set, to generalse onto an American data set despite ethnic differences. In addition, expert graders were surveyed to verify if the DL model was appropriately identifying lesions indicative of nAMD on the OCT scans. Methods Model development data set—12 247 OCT scans from South Korea; external validation data set—91 509 OCT scans from Washington, USA. In both data sets, normal eyes or eyes with nAMD were included. After internal testing, the algorithm was sent to the University of Washington, USA, for external validation. Area under the receiver operating characteristic curve (AUC) and precision–recall curve (AUPRC) were calculated. For model explanation, saliency maps were generated using Guided GradCAM. Results On external validation, AUC and AUPRC remained high at 0.952 (95% CI 0.942 to 0.962) and 0.891 (95% CI 0.875 to 0.908) at the individual level. Saliency maps showed that in normal OCT scans, the fovea was the main area of interest; in nAMD OCT scans, the appropriate pathological features were areas of model interest. Survey of 10 retina specialists confirmed this. Conclusion Our DL algorithm exhibited high performance for nAMD identification in a Korean population, and generalised well to an ethnically distinct, American population. The model correctly focused on the differences within the macular area to extract features associated with nAMD.
19 May 2020 in British Journal of Ophthalmology
Background To investigate the longitudinal associations between person-level and area-level socioeconomic status (PLSES and ALSES, respectively) with diabetic retinopathy (DR) and visual impairment (VI) in Asians with diabetes mellitus (DM). Methods In this population-based cohort study, we included 468 (39.4%) Malays and 721 (60.6%) Indians with DM, with a mean age (SD) of 58.9 (9.1) years; 50.6% were female and the mean follow-up duration was 6.2 (0.9) years. Individual PLSES parameters (education, monthly income and housing type) were quantified using questionnaires. ALSES was assessed using the Socioeconomic Disadvantage Index derived from Singapore’s 2010 areal census (higher scores indicate greater disadvantage). Incident DR and VI were defined as absent at baseline but present at follow-up, while DR and VI progression were defined as a ≥1 step increase in severity category at follow-up. Modified Poisson regression analysis was used to determine the associations of PLSES and ALSES with incidence and progression of DR and VI, adjusting for relevant confounders. Results In multivariable models, per SD increase in ALSES score was associated with greater DR incidence (risk ratio (95% CI) 1.27 (1.13 to 1.44)), DR progression (1.10 (1.00 to 1.20)) and VI incidence (1.10 (1.04 to 1.16)), while lower PLSES variables were associated with increased DR (low income: 1.68 (1.21 to 2.34)) and VI (low income: 1.44 (1.13 to 1.83); ≤4 room housing: 2.00 (1.57 to 2.54)) incidence. Conclusions We found that both PLSES and ALSES variables were independently associated with DR incidence, progression and associated vision loss in Asians. Novel intervention strategies targeted at low socioeconomic status communities to decrease rates of DR and VI are warranted.
26 February 2021 in British Journal of Ophthalmology
Background We examined the associations between the 6-year incidence of age-related macular degeneration (AMD) and vision-related quality of life (VRQoL), and the contribution of presenting visual acuity (VA), in an Asian population. Methods Fundus images from the Singapore Chinese Eye Study, a population-based cohort study (baseline: 2009–2011; follow-up: 2015–2017), were graded using a modified Wisconsin age-related maculopathy grading system. Incident AMD was defined as no baseline AMD in both eyes and early/late AMD in the worse eye at follow-up. Presenting VA was assessed using the logarithm of the minimum angle of resolution chart at 4 m under standard lighting conditions with habitual correction. Multiple linear regression models determined the associations between AMD incidence with changes in the Rasch-transformed scores of the Reading, Mobility and Emotional VRQoL domains of the 32-item Impact of Visual Impairment (IVI-32) questionnaire, adjusted for traditional confounders. The contribution of presenting VA to changes in VRQoL was also estimated. Results Of the 2251 participants without AMD at baseline (mean age (SD): 57.7 (9) years, 51.4% women), 101 (4.5%) and 11 (0.5%) developed incident early and late AMD at follow-up, respectively. Incident late AMD was associated with significant 30.3%, 32.5% and 30.9% decrements in Reading, Mobility and Emotional IVI scores, respectively. The contribution of presenting VA ranged between 1.62% and 4.35% of the observed decrements. No significant associations were noted with incident early AMD. Conclusion Incident late AMD had a substantial impact on all aspects of VRQoL, with presenting VA contributing only minimally to this longitudinal relationship.