Existence of a fully unsupervised deep learning skull stripping method without data manipulation

Develop a fully unsupervised deep learning method for skull stripping in brain MRI that does not rely on any data manipulation during training.

Background

The paper surveys unsupervised approaches with potential for generalizability, including denoising and reconstruction-based anomaly detection methods. However, the authors explicitly state that they do not recognize any fully unsupervised deep learning method for skull stripping that avoids data manipulation.

Creating such a method would address a notable gap in the field by reducing reliance on labeled datasets or synthetic manipulation, potentially improving robustness across diverse modalities and species.