Abstract: In an era defined by the convergence of healthcare and technology, the importance of synthetic imaging data in revolutionizing patient care cannot be overstated. This talk focuses on the role of synthetic imaging data within the healthcare landscape. For unstructured data types like images assessing the utility and gaining industry trust can be especially challenging. Thus, we will delve into the methodologies and best practices for assessing the utility of synthetic imaging data, emphasizing its reliability, accuracy, and ethical considerations. Moreover, we will shed light on the indispensable role that transparency, accountability, and rigorous validation play in building industry trust, ensuring that healthcare stakeholders can confidently embrace synthetic imaging data as a powerful tool for advancing patient diagnosis, treatment, and research.
Abstract: Data bias is a prevalent challenge in real world data (RWD), particularly due to gender and racial disparities between data subjects and the populations they represent. When biased data is used in statistical analysis or the training of machine learning models, this can result in biased algorithms and decisions. This presentation will describe how synthetic data generation techniques can be used to simulate additional individuals from under-represented groups. Such augmented data can mitigate bias in statistical models up to a low to medium levels of data bias. This means that the analysis results are similar to the ground truth (unbiased data). When the bias is high or severe, no particular method performs consistently well for bias mitigation. Early results on bias mitigation in oncology clinical trials will also be presented.