TIACentre/resnet50-kather100k

ResNet-50 Trained on Kather100K (via TIAToolbox)

This model is a ResNet-50 convolutional neural network trained on the Kather100K dataset, which comprises 100,000 histological images of human colorectal cancer and healthy tissue. The model was developed using the TIAToolbox, an end-to-end library for advanced tissue image analytics.

Model Details

• Architecture: ResNet-50
• Parameters: 23.6 million
• Input Size: 224 × 224 × 3
• Dataset: Kather100K (also known as NCT-CRC-HE)
• License: Creative Commons Attribution 4.0 International (CC BY 4.0)
• Source: TIAToolbox

Dataset

The Kather100K dataset consists of 100,000 non-overlapping image patches (224 × 224 pixels) extracted from histological images of human colorectal cancer and healthy tissue. The dataset includes nine tissue classes:

1. Tumor epithelium
2. Simple stroma
3. Complex stroma
4. Immune cells
5. Debris
6. Mucus
7. Adipose tissue
8. Background
9. Muscle

Performance

On the Kather100K dataset, the ResNet-50 model achieved an F1-score of 0.989, indicating high accuracy in classifying histological tissue types.

Usage

Inference with TIAToolbox

from tiatoolbox.models.architecture import get_pretrained_model

# Load the pretrained ResNet-50 model trained on Kather100K
model = get_pretrained_model(pretrained_model='resnet50-kather100k')