Ben R Philps, a PhD student at the University of Edinburgh's School of Informatics, has been awarded the prestigious MIUA 2024 Best Paper Award, sponsored by Springer, for his paper on stochastic uncertainty quantification techniques in White Matter Hyperintensity (WMH) segmentation. Image The paper, titled “Stochastic Uncertainty Quantification Techniques Fail to Account for Inter-Analyst Variability in White Matter Hyperintensity Segmentation,” was recognized at the 28th UK Conference on Medical Image Understanding and Analysis (MIUA), held from July 24-26, 2024, at Manchester Metropolitan University. Reflecting on this achievement, Philps stated, “Winning the award was wonderful as it helped gain attention for the work, and I've had a lot of great discussions with people at the conference about the WMH policy framework we introduced and the associated metrics and how to take the analysis further. It was really encouraging to see others find value in the work that we have done and discuss the complexities and nuances of both WMH segmentation and uncertainty quantification!” Philps’ research addresses a critical issue in medical imaging: the variability among analysts in segmenting WMHs, which are key markers of small vessel disease in brain MRI. “The work was looking at whether uncertainty quantification techniques inherently capture the variability amongst raters for white matter hyperintensities,” Philps explained. He introduced the concept of “annotator policy” in his paper, demonstrating how the variability in annotators’ segmentation policies complicates the assessment of segmentation techniques. Philps and his colleagues developed new metrics, Uncertainty Inter-Rater Overlap (UIRO) and Joint Uncertainty Error Overlap (JUEO), to evaluate these techniques. Their findings suggest that current stochastic uncertainty quantification methods, when trained on a single policy, often fail to capture the inherent variability in WMH segmentation, leading to potential overestimations or underestimations of WMH volume. “White Matter Hyperintensities (WMH) are important neuroradiological markers of small vessel disease in brain MRI, with automatic segmentation tasks essential in research and clinical settings to understand their role in individuals’ health,” Philps noted. He highlighted the challenges in accurate WMH segmentation due to the diversity in shape, intensity, size, and location of the hyperintensities, as well as the varying approaches analysts take in segmentation. The research, co-authored by María del C. Valdés Hernández, Susana Muñoz Maniega, Mark Bastin, Eleni Sakka, Úna Clancy, Joanna Wardlaw, and Miguel O. Bernabeu, highlights the complexities of using multiple annotators in a single study and calls for modifications to segmentation tasks and cost functions to better handle these challenges. The MIUA conference serves as a major forum in the UK for researchers across disciplines to discuss advancements in medical image processing and analysis. Philps, B. et al. (2024). Stochastic Uncertainty Quantification Techniques Fail to Account for Inter-analyst Variability in White Matter Hyperintensity Segmentation. In: Yap, M.H., Kendrick, C., Behera, A., Cootes, T., Zwiggelaar, R. (eds) Medical Image Understanding and Analysis. MIUA 2024. Lecture Notes in Computer Science, vol 14859. Springer, Cham. https://doi.org/10.1007/978-3-031-66955-2_3. Publication date 12 Aug, 2024
