Health Citizens – European Research Center

Tag: AI Diagnostics

  • Fairness And Bias Mitigation in AI Models for Diabetes Diagnosis: A Comparative Evaluation of Algorithmic Approaches

    Fairness And Bias Mitigation in AI Models for Diabetes Diagnosis: A Comparative Evaluation of Algorithmic Approaches

    AI models that predict diabetes can produce biased results for certain groups of patients, and several methods have been developed to reduce this. This study compared three of these methods on the same data, a widely used dataset of 768 women aged 21 and over, assessing fairness across age groups.

    Each method had different strengths. The first relied on simple, transparent decision rules, making its predictions easy to interpret, but it had no specific mechanism to ensure fairness. The second was designed to prevent the model from basing its predictions on age, and achieved perfect results on the fairness measures used while keeping accuracy at an acceptable level. The third corrected the imbalance between patients with and without diabetes in the data, which gave the highest overall accuracy but did not ensure fairness.

    The authors conclude that none of the methods achieves the right balance between accuracy, fairness and interpretability on its own, and propose combining several strategies to reduce bias in AI-based diabetes diagnosis.

    Learn more about this study here: https://doi.org/10.54554/jtec.2025.17.03.004

    Reference:

    Mohd Hamdan, M.D.H., Ab Jabal, M.F., Abdul Rahman, S., & Kapi, A.Y. (2025). Fairness And Bias Mitigation in AI Models for Diabetes Diagnosis: A Comparative Evaluation of Algorithmic Approaches. Journal of Telecommunication, Electronic and Computer Engineering, 17(3), 27–34.

  • Mitigating bias in deep learning for diagnosis of coronary artery disease from myocardial perfusion SPECT images

    Mitigating bias in deep learning for diagnosis of coronary artery disease from myocardial perfusion SPECT images

    AI can accurately diagnose coronary artery disease from myocardial perfusion imaging, a type of scan that shows blood flow to the heart muscle. However, these models are often trained on high-risk patients who were referred for invasive tests, which can lead them to overestimate the probability of disease in other patients. This study tested ways of training AI so that the probabilities it estimates match the real probability of disease more closely.

    Deep learning models were trained on 828 patients from three sites and then tested on a separate group of 511 patients, including 179 with a low likelihood of disease. The best results came from a model trained with additional examples of patients without coronary artery disease and with little or no sign of reduced blood flow on the scan. It produced more accurate probability estimates than the model trained without these additional cases, and the improvement was most pronounced in women. It was also better at identifying disease than standard measures taken from the scans.

    The authors conclude that adding more low-risk patients during training can improve how accurately AI estimates the probability of disease. This matters especially for lower-risk patients and for women, for whom overestimated probabilities could affect decisions about further tests and treatment.

    Learn more about this study here: https://doi.org/10.1007/s00259-022-05972-w

    Reference:

    Miller, R.J.H., Singh, A., Otaki, Y., Tamarappoo, B.K., Kavanagh, P., Parekh, T., Hu, L.-H., Gransar, H., Sharir, T., Einstein, A.J., Fish, M.B., Ruddy, T.D., Kaufmann, P.A., Sinusas, A.J., Miller, E.J., Bateman, T.M., Dorbala, S., Di Carli, M.F., Liang, J.X., … Slomka, P.J. (2023). Mitigating bias in deep learning for diagnosis of coronary artery disease from myocardial perfusion SPECT images. European Journal of Nuclear Medicine and Molecular Imaging, 50(2), 387–397.

  • Addressing Artificial Intelligence Bias in Retinal Diagnostics

    Addressing Artificial Intelligence Bias in Retinal Diagnostics

    Diabetic retinopathy, damage to the retina caused by diabetes, is one of the conditions in which AI is used to support diagnosis by analysing images of the back of the eye. This study looked at what happens when the data used to train these systems is unbalanced, and tested a way of reducing the resulting bias.

    The researchers used a public dataset of retinal images to which clinicians had added skin-tone labels. To recreate a situation of imbalance, they removed from the training data the images of people with darker skin who had diabetic retinopathy requiring referral to a specialist, while keeping these images in the test data. The standard system was 73.0% accurate for people with lighter skin and 60.5% for people with darker skin, a gap of 12.5 percentage points.

    They then trained the system with additional synthetic retinal images created by generative AI models. Accuracy became 72.0% for people with lighter skin and 71.5% for people with darker skin, reducing the gap to 0.5 points. The authors conclude that unbalanced training data can lead to unequal accuracy across groups, and that synthetic images may help reduce this bias in the diagnosis of diabetic retinopathy and possibly in other AI systems used in eye care.

    Learn more about this study here: https://doi.org/10.1167/tvst.10.2.13

    Reference:

    Burlina, P., Joshi, N., Paul, W., Pacheco, K.D., & Bressler, N.M. (2021). Addressing Artificial Intelligence Bias in Retinal Diagnostics. Translational Vision Science & Technology, 10(2), 13.