PyHistopathology¶

PyHistopathology is a Python API for pre-processing Whole-Slide Images for use in a machine learning algorithm.

Features¶

Reading Whole-Slide Images with WSI_Scanning.readWSI()

Input: WSI path or directory
Output: Numpy array of WSI with data type int32

Denoising Whole-Slide Images with Denoising.denoising()

Input: WSI Path or directory
Output: Numpy array of WSI Image (After denoising) with dtype int32

Patch Extraction for Whole-Slide Images with Extractingpatches.extractingPatches()

Input: WSI Path or directory
Output: Fills up the outputpath with patches extracted from the WSI

Table of Contents¶

Reading¶

Use WSI_Scanning.readWSI() to read Whole-Slide Images

Usage¶

readWSI(WSI_path, magnification_level, annotation_file, annonated_level)

Arguments¶

WSI_path¶

Path or location of WSI.

magnification_level¶

Level of zoom, for example 40, 20, 10, or 5. Default magnification level is 20. - Note: if magnification 40x for max zoom level of 20x image an error will be raised.

annotation_file¶

Default annotation = None. If annotation are available in a xml file, set annotation_file to be the xml file path.

annonated_level¶

If annotation is given then set annotated_level equal to the z-axis of the annotations. Default annotatedlevel is 0.

Return Type¶

Numpy array of WSI Image (After denoising) with dtype int32

Example¶

from WSI_Preprocessing.Preprocessing import WSI_Scanning
import cv2
img,slide_dim = WSI_Scanning.readWSI("example.svs")
cv2.imwrite("example.png",img)

Denoising¶

Use Denoising.denoising() to remove stains, folds and other background noise in Whole-Slide Images

Usage¶

denoising(inputsvs, magnification, filtering, patch_size, upperlimit, lowerlimit, red_value, green_value, blue_value)

Arguments¶

inputsvs¶

Path or location of WSI.

magnification¶

Level of zoom, for example 40, 20, 10, or 5. Default magnification level is 20. - Note: if magnification 40x for max zoom level of 20x image an error will be raised.

filtering¶

GuassianBlur, RGBThersholding, or None

GuassianBlur: Homogeneity calculations based on image smoothing and Gaussian blur equations. We compute sum of square differences between two consecutive Gaussian blurred images as score for homogeneity.

Upper limit: Upper threshold of homogeneity score. Default value is 9500 with kernel size of 1111
Lower limit: lower threshold of homogeneity score. default value is 1500 with kernel size of 1111
Patch size: Not significant parameters for GuassianBlur filtering

RGBThersholding: Validated patches based on RGB values of patches

red_value: Red threshold
green-value: Green threshold
blue_value: Blue Threshold

None: Only removes Background

Note that our default is GuassianBlur technique. GuassianBlur is highly effective and requires more computational power (RAM). RGBThersholding is less effective which needs less computational power

Return Type¶

Numpy array of WSI Image (After denoising) with dtype int32

Extracting¶

Use Extractingpatches.extractingPatches() to extract patches from Whole-Slide Images

Usage¶

extractingPatches(inputsvs, outputpath, magnification, patch_extraction_creatia, number_of_patches, filtering, patch_size, upperlimit, lowerlimit, red_value, green_value, blue_value, Annotation, Annotationlevel, Requiredlevel, reconstructionimagepath)

Arguments¶

inputsvs¶

Path or location of WSI.

magnification¶

Level of zoom, for example 40, 20, 10, or 5. Default magnification level is 20. - Note: if magnification 40x for max zoom level of 20x image an error will be raised.

filtering¶

GuassianBlur, RGBThersholding, or None

GuassianBlur: Homogeneity calculations based on image smoothing and Gaussian blur equations. We compute sum of square differences between two consecutive Gaussian blurred images as score for homogeneity.

Upper limit: Upper threshold of homogeneity score. Default value is 9500 with kernel size of 1111
Lower limit: lower threshold of homogeneity score. default value is 1500 with kernel size of 1111
Patch size: Not significant parameters for GuassianBlur filtering

RGBThersholding: Validated patches based on RGB values of patches

red_value: Red threshold
green-value: Green threshold
blue_value: Blue Threshold

None: Only removes Background

Note that our default is GuassianBlur technique. GuassianBlur is highly effective and requires more computational power (RAM). RGBThersholding is less effective which needs less computational power

patch_extraction_creatia¶

random, or None. Default is None. For extracting a fixed number of patches for WSI we can use random.

number_of_patches¶

Default number of patches is 2000

outputpath¶

Folder to store the extracted patches

reconstructionimagepath¶

If you want to compare the patches with WSI we can mention the reconstructionimagepath. Default is None. Note: it only works with patch_extraction_creatia = None.

Return Type¶

None, fills up output path with images directly instead of returning a Numpy array.