Abstract
Atlas-building from population data is widely used in medical imaging. However, the emphasis of atlas-building approaches is typically to compute a mean / median shape or image based on population data. In this work, we focus on the statistical characterization of the population data, once spatial alignment has been achieved. We introduce and propose the use of the weighted functional boxplot. This allows the generalization of concepts such as the median, percentiles, or outliers to spaces where the data objects are functions, shapes, or images, and allows spatio-temporal atlas-building based on kernel regression. In our experiments, we demonstrate the utility of the approach to construct statistical atlases for pediatric upper airways and corpora callosa revealing their growth patterns. Furthermore, we show how such atlas information can be used to assess the effect of airway surgery in children.
Introduction
In medical imaging, researchers often build "atlases"—representative templates of a population. Traditionally, these atlases focus on the mean or median shape. However, this "average-only" approach has two major flaws: it discards crucial information about the population's diversity (such as percentiles and outliers) and it struggles with data sparsity. When trying to see how an organ changes with age, researchers often "bin" data. If there are only a few subjects at a specific age, the statistical estimate becomes unreliable.
Methods: Generalizing the Boxplot
This paper introduces the weighted functional boxplot, a novel statistical tool that generalizes traditional boxplots to complex data like curves, shapes, and images. The authors extend the concept of Band-Depth—a measure of how "central" a curve is within a bundle of curves—into a Weighted Band-Depth (WBD).
By incorporating kernel-based weighting (using a Gaussian function), the algorithm can build a "sliding window" atlas. To build an atlas for a 5-year-old, the algorithm looks at the whole dataset but gives more weight to subjects closer to that target age. Unlike "point-wise" statistics, which calculate the median at every single coordinate independently, functional boxplots treat the entire curve or shape as a single unit. This ensures the "median" is an actual representative shape from the dataset, rather than a non-physical average.
Key Insights and Results
The method was evaluated on synthetic sine waves, pediatric upper airways, and the corpus callosum. A major advantage of this approach is its ability to overcome data sparsity. By using a weighted approach, the authors can build a robust atlas even for ages where they have very few subjects, as neighboring ages provide enough context to fill the gaps.
In tests on real airway data, point-wise confidence regions only contained 5 of 44 subjects, whereas the functional boxplot correctly captured 50% of the population within its 50% central region. Furthermore, the 50% central region (the "box" of the boxplot) provides a rigorous definition of what a "normal" anatomy looks like at a specific age, which is invaluable for clinical diagnosis.
Clinical Application
The authors applied the atlas to evaluate a child with Subglottic Stenosis (SGS), an airway narrowing condition. Before surgery, the child's airway curve was a clear outlier, falling far outside the atlas "fences." Post-surgery, the curve moved inside the "maximal non-outlying envelope," quantitatively proving the surgery's success in returning the airway to a normal range.
Conclusion and Future Work
The weighted functional boxplot offers a robust framework for spatio-temporal atlas construction. While it requires high-quality spatial alignment of data, it preserves the natural geometry of anatomical structures. Future work may involve building these order statistics on low-dimensional manifolds to handle extremely high-dimensional image data more efficiently.
Summary updated