You can deploy this model and start earning money today!

drneumann/
monusac-run9-segmentation

Name: monusac-run9-segmentation
Availability: InStock

mit

Views

No views yet

MoNuSAC Nucleus Segmentation – Run9 Checkpoint

Overview

This repository hosts the trained checkpoint for a multi-head U-Net model trained on the MoNuSAC 2020 dataset for nucleus segmentation and classification in H&E histopathology images.

The model jointly predicts:

5-class semantic segmentation (background + 4 nucleus types)
Ternary structure map (background / inside / boundary)

Dataset

MoNuSAC 2020
Multi-organ H&E-stained histopathology
40x magnification

Architecture

Encoder: ResNet (via segmentation_models_pytorch)
Decoder: U-Net style
Dual output heads:
- Semantic segmentation (5 classes)
- Ternary boundary segmentation (3 classes)

Validation Performance (Best Checkpoint)

Task	Metric	Value
Semantic	mIoU (macro)	0.756
Semantic	mIoU (micro)	0.856
Ternary	Dice (inside)	0.673
Ternary	Dice (boundary)	0.608
Ternary	mIoU (micro)	0.801

Usage

The checkpoint can be loaded using PyTorch:

python

import torch
ckpt = torch.load("run9__best.pt", map_location="cpu")

drneumann/monusac-run9-segmentation