Figure 1.

Many modern algorithms in medical image processing are developed using magnetic resonance (MR) scans with high resolution and SNR due to research conditions. However, it is unclear how these algorithms translate to noisier, more heterogeneous data acquired during clinical practice. Here, we compare a research scan from the OASIS dataset to two clinically acquired brain MR imaging: a pediatric subject and an adult subject. These images highlight the heterogeneity in clinical data due to differences in anatomy, contrast, or noise. In this work, we propose a solution to generalize an existing segmentation algorithm to these circumstances.