Dr. Bruce Hashemian
Dr. Bruce Hashemian
**Is your feature request related to a problem? Please describe.** The inference in pathology pipelines is usually based on patches and to create the whole picture these patches need to...
Often times MONAI users need to collate the patches extracted from the same WSI. Currently, we are relying on the file path to perform such operation but similar to OpenSlide...
Fixes #6138 ### Description This PR makes the batch sampling of `PatchInferer` multi-threaded so that WSI inference (using WSI splitters) can leverage this to enhance performance. ### Types of changes...
Stain normalization requires a stain matrix that are particular to an image or a set of image acquired under the same condition and settings. In MONAI pathology, we have this...
HoVerNet post-processing is an essential part of the hoverne pipeline but it is taking too much time (way more than inference itself). @JHancox has some experiences to make it run...
`SlidingWindowSplitter` currently supports already loaded images into tensors or read whole slide images using `WSIReader`. Using current API (with backward compatiblity), we should be able to add support for any...
CuCIM is a dependency of MONAI for efficiently loading whole slide images and also some GPU accelerated transforms that are being accessed via `CuCIM` transform in MONAI. CuCIM has dropped...
One strategy to merge patches that are the outputs of the inference model (e.g., a probability map) is to take the minimum or maximum of the prediction at the overlapping...
One strategy to merge patches that we are already supporting in `SligingWindowInferer` is to use Gaussian weights to blend the patches together. To support this feature in the new design,...