Brief news summary

A recent study in Nature Medicine revealed that AI machine learning models used in medical imaging exhibit biases comparable to human biases. Although these models are highly proficient at predicting race and gender, they may not perform as effectively in disease diagnosis. This raises concerns about the increasing use of AI in medical imaging. Biases in AI are influenced by various factors, including the quantity and quality of training data, data structure, and human biases. To address this issue, regular evaluation and auditing of AI models are essential to mitigate biases. The study found that while the overall performance of the AI models was satisfactory, there were variations in prediction accuracies based on gender and race. The study recommends implementing specific techniques to reduce biases and stresses the importance of evaluating models using patients who are similar to those on whom they were trained. To achieve accurate results across diverse demographics, hospitals using AI models should thoroughly evaluate algorithms with their own patient data, particularly when utilizing external data.