CrossMoDA challenge: Progress in unsupervised cross-modality segmentation of vestibular schwannoma and cochlea

Adding heterogeneous training data reduced segmentation outliers and improved model generalization for vestibular schwannoma, but cochlea segmentation accuracy declined in the 2023 edition of the crossMoDA challenge.

• The 2023 winning approach generalized well to earlier homogeneous test sets, supporting use of heterogeneous training data.
• Average cochlear Dice scores declined in 2023, possibly due to increased dataset heterogeneity, resolution variability, and added segmentation complexity; the cause remains uncertain.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.