Pathology_DetectInk​edBoundaries

automated detection and digital mapping of inked markings on pathology specimens

DetectInkedBoundaries

Stephanie Harmon
[e-mail]
Leidos Biomedical Research
Molecular Imaging Branch, National Cancer Institute
National Institutes of Health
December 2018

---DESCRIPTION --------------------------------------------------------------------------------------------------------------
This program takes in digital pathology (.czi/.svs) files with inked markings, automatically detects boundaries of ink, provides user interface for editing of detected boundaries, and writes to annotation file (.cz/.xml).

(Optional) Additionally, users can provide corresponding un-inked digital imaging, which will be automatically registered to inked imaging and annotations saved

(Optional - NEW 1/7/2019) users can elect to manually outline inked borders, which are then automatically registered to un-inked imaging (if provided) and annotations automatically saved. This may come in use if the 'auto' option fails to accurately detect boundaries

Annotations will be saved to the highest resolution image contained in the digital stack (i.e. 40x if full digital image) provided at input.

This program utilizes other freely available toolboxes, see dependencies

--- WORKFLOW --------------------------------------------------------------------------------------------------------------

1. read in image stack, find lowest magnification ratio (smallest
sampling of full-res image)
2. Option 'auto' ink detection/outlining
2a. deconvolve RGB H&E image using Khan et al Classifier
2b. Make a mask of tissue sample using H channel
2c. Use k-means clustering to detect inked markings from
remaining channels within tissue sample
2d. Grow-shrink morhological operations to determine ROIs
2e. Initiate user interface for accept/reject of all
proposed ROIs
2f. Initiate user interface to allow editing of accepted ROIs
Option 'manual' ink outlining
3a. user prompted for number of ROIs
3b. initiate user interface for manual outlining
3c. after each ROI is outlined, user should close figure
3. Correlation-based image registration of inked and un-inked
4. Write annotation file
(.xml if supplied .svs or .cz if supplied .czi)

---DEPENDENCIES -----------------------------------------------------------------------------------------------------------

- local installation of MATLAB 2018b

- MATLAB Image Processing Toolbox

- Bio-Formats toolbox for MATLAB
https://docs.openmicroscopy.org/bio-formats/5.3.4/users/matlab/index.html

- color decon toolbox from Khan et al
https://github.com/lun5/color-deconvolution/tree/master/stain_normalisation_toolbox

- polygon decimate function from
https://www.mathworks.com/matlabcentral/fileexchange/34639-decimate-polygon

--- USAGE --------------------------------------------------------------------------------------------------------------

DetectInkedBoundaries('--Marked','/example/path/to/marked.svs','--Unmarked', '/example/path/to/unmarked.svs','--Method','auto')

input:
1. (required) digital image in CZI or SVS format with inked markings
2. (optional) digital image of specimen with removed markings, if provided, will register low res
digital images and register digital markings to 'clean' specimen
3. (optional) method of annotation. default = 'auto'. users who wish to outline images
themselves should use 'manual'

--- TIPS AND TRICKS -------------------------------------------------------------------------------------------------------

- Points can be added and deleted from ROIs after accept/reject stage, this stage is meant to
exclude false positives arising from other inked notes/markings (i.e. "EPE", arrows, etc).

- If several proposed ROIs are part of the same region, simply add more points to join ROIs as part
of Workflow Step 2f. A closing operation is performed that will merge all overlapping ROIs.

- If an ROI is open-ended (i.e. not closed circle), the tool will encompass only ink. To expand to other
regions, all add and/or drag points to encompass the full ROI area