Style-Aware Blending and Prototype-Based Cross-Contrast Consistency for Semi-Supervised Medical Image Segmentation

Chaowei Chen¹,Xiang Zhang¹,Honglie Guo¹, Shunfang Wang^1,2,*

¹School of Information Science and Engineering, Yunnan University, Kunming, 650504, Yunnan, China

²Yunnan Key Laboratory of Intelligent Systems and Computing, Yunnan University, Kunming, 650504, Yunnan, China

Paper: (arXiv 2507.20729)

Abstract

Weak-strong consistency learning strategies are widely employed in semi-supervised medical image segmentation to train models by leveraging limited labeled data and enforcing weak-to-strong consistency. However, existing methods primarily focus on designing and combining various perturbation schemes, overlooking the inherent potential and limitations within the framework itself. In this paper, we first identify two critical deficiencies: (1) separated training data streams, which lead to confirmation bias dominated by the labeled stream; and (2) incomplete utilization of supervisory information, which limits exploration of strong-to-weak consistency. To tackle these challenges, we propose a style-aware blending and prototype-based cross-contrast consistency learning framework. Specifically, inspired by the empirical observation that the distribution mismatch between labeled and unlabeled data can be characterized by statistical moments, we design a style-guided distribution blending module to break the independent training data streams. Meanwhile, considering the potential noise in strong pseudo-labels, we introduce a prototype-based cross-contrast strategy to encourage the model to learn informative supervisory signals from both weak-to-strong and strong-to-weak predictions, while mitigating the adverse effects of noise. Experimental results demonstrate the effectiveness and superiority of our framework across multiple medical segmentation benchmarks under various semi-supervised settings.

Overview

We propose a style-aware blending and prototype-based cross-contrast consistency framework for semi-supervised medical image segmentation. The method addresses distribution mismatch between labeled and unlabeled data while enabling bidirectional consistency learning.

Main Results

Synapse Multi-Organ Dataset

Method	Labeled	DSC	ASD	Aorta	GB	KL	KR	Liver	PC	SP	SM
Ours (MT)	1 (5%)	60.00	19.73	79.53	40.52	68.59	63.05	89.80	20.70	78.38	39.00
Ours (BCP)	1 (5%)	61.87	18.81	80.07	38.78	68.30	63.12	91.60	24.54	80.65	47.93
Ours (MT)	4 (10%)	65.81	15.88	80.77	45.42	76.39	76.45	90.10	30.14	75.84	51.36
Ours (BCP)	4 (10%)	66.59	23.52	81.79	40.09	76.80	68.64	90.43	38.80	75.79	60.39

ACDC Cardiac Dataset

Method	Labeled	DSC	ASD	Myo	LV	RV
Ours (MT)	3 (5%)	87.31	1.10	85.60	84.90	91.42
Ours (BCP)	3 (5%)	88.60	0.61	87.36	85.51	92.93
Ours (MT)	7 (10%)	89.27	0.63	88.35	86.63	92.82
Ours (BCP)	7 (10%)	89.80	0.95	90.36	86.10	92.94

Installation

We recommend the following platforms:

Python 3.8+ / PyTorch 2.0.0+ / CUDA 11.8+

Install the necessary packages:

pip install -r requirements.txt

Set the dataset path environment variable:

export DATASET_HOME=/path/to/your/datasets

Dataset Preparation

ACDC Cardiac Dataset

Download the preprocessed ACDC dataset from SSL4MIS and extract and organize the data:

$DATASET_HOME/
└── ACDC/
    ├── data/
    │   ├── slices/*.h5
    │   └── *.h5
    └── train_slices.list
    └── val.list
    └── test.list

Synapse Multi-Organ Dataset

Sign up at the official Synapse website or download the dataset using the preprocessed data. Split the dataset according to the 5% and 10% protocols and organize it in a structure similar to the ACDC dataset.

Training

Mean Teacher Framework

# Synapse dataset
python train_Ours_MT.py --name Synapse --labelnum 5   # 5% labeled
python train_Ours_MT.py --name Synapse --labelnum 10  # 10% labeled

# ACDC dataset
python train_Ours_MT.py --name ACDC --labelnum 5      # 5% labeled
python train_Ours_MT.py --name ACDC --labelnum 10     # 10% labeled

BCP Framework

# Synapse dataset
python train_Ours_BCP.py --name Synapse --labelnum 5   # 5% labeled
python train_Ours_BCP.py --name Synapse --labelnum 10  # 10% labeled

# ACDC dataset
python train_Ours_BCP.py --name ACDC --labelnum 5      # 5% labeled
python train_Ours_BCP.py --name ACDC --labelnum 10     # 10% labeled

Testing

# Testing on Synapse dataset
bash test_Synapse.sh

# Testing on ACDC dataset
bash test_ACDC.sh

Acknowledgments

We thank the authors of SSL4MIS, TransUNet, and BCP for making their valuable code & data publicly available.