Skip to main content

​
Introduction

DigiPathAI provides a Python API for programmatic access to its tissue segmentation capabilities. The API allows you to integrate whole slide image (WSI) analysis directly into your workflows without using the web interface.

​
Import Structure

The main functions are available from the DigiPathAI.Segmentation module: 
from DigiPathAI.Segmentation import getSegmentation, get_prediction

​
Main Entry Points

DigiPathAI provides two primary functions for tissue analysis:

​
getSegmentation

The high-level function for complete tissue segmentation. This is the recommended entry point for most use cases. 
prediction = getSegmentation(
    img_path,
    patch_size=256,
    stride_size=128,
    batch_size=32,
    quick=True,
    mode='colon'
)
Use when: You need a complete end-to-end segmentation with model loading, inference, and post-processing.

​
get_prediction

A lower-level function for patch-based prediction with custom models. Provides more control over the inference process. 
img, probs_map = get_prediction(
    wsi_path,
    batch_size=64,
    models=custom_models,
    patch_size=256,
    stride_size=256
)
Use when: You need fine-grained control over model selection or want to use custom trained models.

​
Basic Usage Pattern

Here’s a typical workflow using the Python API: 
from DigiPathAI.Segmentation import getSegmentation

# Perform tissue segmentation
prediction = getSegmentation(
    img_path='path/to/slide.tiff',
    patch_size=256,
    stride_size=128,
    batch_size=32,
    quick=True,
    tta_list=None,
    crf=False,
    mode='colon'
)

# prediction is a numpy array containing the segmentation mask
print(f"Prediction shape: {prediction.shape}")
print(f"Unique values: {np.unique(prediction)}")
The quick=True parameter uses a single model for faster inference. Set quick=False to ensemble 4 different models for higher accuracy.

​
Return Types and Data Structures

​
getSegmentation Return Value

Returns a numpy array containing the binary segmentation mask:
  • Shape: (height, width) matching the WSI dimensions at the highest resolution
  • Data type: float32 or uint8
  • Values:
    • 255 = tissue (positive prediction)
    • 0 = background (negative prediction)
prediction = getSegmentation(img_path, mode='colon')
# prediction.shape = (10000, 15000)  # Example dimensions
# prediction.dtype = numpy.float32
# Unique values: [0, 255]

​
get_prediction Return Value

Returns a tuple of (image, probs_map): 
img, probs_map = get_prediction(wsi_path, models=models)
Components:
  • img: The original WSI as a numpy array
  • probs_map: Dictionary containing prediction statistics
    • probs_map['mean']: Mean probability map across all models/augmentations
    • probs_map['var']: Variance (uncertainty) map
​
probs_map
dict
Dictionary containing probability maps with mean and variance
​
mean
numpy.memmap
Mean probability map. Values range from 0.0 to 1.0, where higher values indicate stronger tissue predictions.
​
var
numpy.memmap
Variance map representing prediction uncertainty. Higher values indicate less confident predictions.

​
Performance Considerations

Processing whole slide images requires significant GPU memory. Monitor your GPU usage and adjust batch_size accordingly.
Key parameters affecting performance:
  • batch_size: Higher values = faster processing but more GPU memory
  • patch_size: Larger patches = fewer iterations but more memory per patch
  • stride_size: Smaller strides = more overlap and better quality but slower processing
  • quick: True = single model (fast), False = 4-model ensemble (slow but accurate)

​
Next Steps

getSegmentation Function

Detailed documentation for the main segmentation function

Parameter Reference

Complete reference for all function parameters

Build docs developers (and LLMs) love

Get started for freeTalk to us