Skip to content

Commit 50fbb66

Browse files
committed
updated readme
1 parent 67af4c1 commit 50fbb66

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

README.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -13,7 +13,7 @@
1313
Pythostitcher is a Python implementation of the [AutoStitcher](https://www.nature.com/articles/srep29906) software which stitches prostate histopathology images into an artifical whole mount image. Although the general stitching mechanism of Pythostitcher is similar to AutoStitcher, PythoStitcher offers several new features, most notably the ability to reconstruct a full resolution (0.25 µm) whole-mount. Furthermore, PythoStitcher is implemented in Python and significantly faster compared to the [original implementation in Matlab](https://engineering.case.edu/centers/ccipd/content/software). In addition, Pythostitcher offers several new advanced features such as 1) automatically finding the optimal fragment configuration, 2) providing a smooth overlap between overlapping parts with gradient alpha blending and 3) providing support for a varying number of tissue fragments.
1414

1515
## Algorithm input
16-
The input for Pythostitcher consists of either two or four high resolution histopathology images and the corresponding tissue masks. These must both be pyramidal files and consist of incrementally downsampled resolutions (preferable .mrxs or .tif format). These tissue masks can be generated by your tissue segmentation algorithm of choice, in the provided sample data we make use of the algorithm from [Bándi et al](https://pubmed.ncbi.nlm.nih.gov/31871843/). Although not required, the user may specify the desired location of the tissue fragments in the final reconstruction. In the case of a prostatectomy cross-section which consists of two fragments, these locations would usually be left and right. Currently, supported locations are ['right', 'left', 'top', 'bottom']. If these locations are not specified, PythoStitcher will automatically figure out the most suitable way to reconstruct the whole-mount.
16+
The input for Pythostitcher consists of either two or four high resolution histopathology images and the corresponding tissue masks. These must both be pyramidal files and consist of incrementally downsampled resolutions (preferable .mrxs or .tif format). These tissue masks can be generated by your tissue segmentation algorithm of choice, in the provided sample data we make use of the algorithm from [Bándi et al](https://pubmed.ncbi.nlm.nih.gov/31871843/). Although not required, the user may specify the desired location of the tissue fragments in the final reconstruction. In the case of a prostatectomy cross-section which consists of two fragments, these locations would usually be left and right. Currently, supported locations are ['right', 'left', 'top', 'bottom']. If these locations are not specified, PythoStitcher will automatically figure out the most suitable way to reconstruct the whole-mount. This will be performed by using [JigsawNet](https://github.com/Lecanyu/JigsawNet), a CNN trained to identify adjacent fragments in jigsaw puzzles.
1717

1818
## Algorithm details
1919
After preprocessing the input images, Pythostitcher will perform an automated stitch edge detection and compute a rough initial alignment on heavily downsampled versions of the input images. This initial alignment is then iteratively refined using images with an increasingly finer resolution. This refinement is performed by a genetic algorithm, which aims to minimize the average distance between the stitch edges of adjacent fragments. One of the strengths of Pythostitcher is that the optimal alignment between fragments can be scaled linearly for finer resolutions. Hence, when a satisfactory alignment is achieved on a lower resolution, this alignment can be scaled up linearly to compute the full resolution stitched result. This greatly improves computational overhead, since the full resolution images can be up to ~100k pixels in height/width, making any direct image processing infeasible on a regular clinical workstation.

0 commit comments

Comments
 (0)