stlearn.pp.extract_feature(adata: AnnData, cnn_base: Literal['resnet50', 'vgg16', 'inception_v3', 'xception'] = 'resnet50', n_components: int = 50, verbose: bool = False, copy: bool = False, seeds: int = 1) Optional[AnnData][source]

Extract latent morphological features from H&E images using pre-trained convolutional neural network base

  • adata – Annotated data matrix.

  • cnn_base – Established convolutional neural network bases choose one from [‘resnet50’, ‘vgg16’, ‘inception_v3’, ‘xception’]

  • n_components – Number of principal components to compute for latent morphological features

  • verbose – Verbose output

  • copy – Return a copy instead of writing to adata.

  • seeds – Fix random state


  • Depending on copy, returns or updates adata with the following fields.

  • **X_morphology** (adata.obsm field) – Dimension reduced latent morphological features.