Skip to main content

​
Overview

DigiPathAI provides specialized models for liver cancer segmentation, trained on the PAIP (Pathology AI Platform) dataset. These models detect and segment viable tumor regions in hepatocellular carcinoma whole slide images.

​
PAIP Dataset

The PAIP dataset contains annotated liver cancer histopathology slides from hepatocellular carcinoma cases. The model ensemble includes:

Inception

Multi-scale feature extraction for liver tissue patterns

DenseNet

Dense connections for robust feature learning

DeepLabV3

Accurate tumor boundary delineation

​
Model Details

The liver cancer models are automatically managed by DigiPathAI: 
# From Segmentation.py lines 248-262
elif mode == 'liver':
    path = os.path.join(home, '.DigiPathAI/paip_models')
    if (not os.path.exists(os.path.join(path, 'paip_inception.h5'))) or \
        (not os.path.exists(os.path.join(path, 'paip_deeplabv3.h5'))) or \
        (not os.path.exists(os.path.join(path, 'paip_densenet.h5'))):
        if status is not None: status['status'] = "Downloading Trained Models"
        download_paip() 
        model_path_inception = os.path.join(path, 'paip_inception.h5')
        model_path_deeplabv3 = os.path.join(path, 'paip_deeplabv3.h5')
        model_path_densenet  = os.path.join(path, 'paip_densenet.h5')
    else :
        if status is not None: status['status'] = "Found Trained Models, Skipping download"
        model_path_inception = os.path.join(path, 'paip_inception.h5')
        model_path_deeplabv3 = os.path.join(path, 'paip_deeplabv3.h5')
        model_path_densenet  = os.path.join(path, 'paip_densenet.h5')
Models are stored in ~/.DigiPathAI/paip_models/ and include:
  • paip_inception.h5
  • paip_deeplabv3.h5
  • paip_densenet.h5

​
Usage Example

​
Basic Liver Segmentation

from DigiPathAI.Segmentation import getSegmentation

# Segment liver cancer tissue
prediction = getSegmentation(
    img_path='liver_slide.tiff',
    mode='liver',
    patch_size=256,
    stride_size=128,
    batch_size=32,
    quick=False  # Use ensemble of all three models
)

​
Quick Mode (Single Model)

# Use only DeepLabV3 for faster inference
prediction = getSegmentation(
    img_path='liver_slide.tiff',
    mode='liver',
    model='deeplabv3',
    quick=True
)

​
With Custom Paths

# Specify output paths for results
prediction = getSegmentation(
    img_path='liver_slide.tiff',
    mode='liver',
    probs_path='./results/liver_probs.tiff',
    mask_path='./results/liver_mask.tiff',
    uncertainty_path='./results/liver_uncertainty.tiff',
    quick=False
)
For hepatocellular carcinoma analysis, the ensemble mode (quick=False) provides more reliable tumor boundary detection and better handles tissue heterogeneity.

​
Parameters

ParameterDescriptionDefaultLiver-Specific Notes
modeTissue type-Must be set to 'liver'
modelArchitecture choice'dense'Options: ‘dense’, ‘inception’, ‘deeplabv3’
quickSingle vs ensembleTrueSet to False for ensemble
patch_sizeInference patch size256Recommended: 256 for liver
stride_sizeSliding window stride128Smaller values = more overlap
batch_sizeBatch size for inference32Adjust based on GPU memory

​
Output

The segmentation returns three files for comprehensive tumor analysis:
  1. Probability Map - Continuous probability values indicating tumor likelihood
  2. Binary Mask - Thresholded segmentation (threshold=0.3) for viable tumor regions
  3. Uncertainty Map - Model variance for assessing prediction confidence
# All outputs are saved as pyramidal TIFF files
# Optimized for viewing in digital pathology viewers
prediction = getSegmentation(
    img_path='liver_slide.tiff',
    mode='liver',
    probs_path='./liver_probability.tiff',
    mask_path='./liver_mask.tiff',
    uncertainty_path='./liver_uncertainty.tiff'
)

​
Clinical Applications

  • Viable Tumor Quantification: Measure tumor area percentage
  • Treatment Response: Compare pre/post-treatment tumor regions
  • Research: Correlate tumor morphology with outcomes
  • Quality Control: Use uncertainty maps to identify challenging regions

Build docs developers (and LLMs) love

Get started for freeTalk to us