Interactive Medical Image Segmentation: A Benchmark Dataset and Baseline

A Pytorch implement of medical image segmentation U-shape architecture benchmarks

AMOS: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation

With recent advances in machine learning, semantic segmentation algorithms are becoming increasingly general-purpose and translatable to unseen tasks. Many key algorithmic advances in the field of medical imaging are commonly validated on a small number of tasks, limiting our understanding of the generalizability of the proposed contributions. A model which works out-of-the-box on many tasks, in the spirit of AutoML (Automated Machine Learning), would have a tremendous impact on healthcare. The field of medical imaging is also missing a fully open source and comprehensive benchmark for general-purpose algorithmic validation and testing covering a large span of challenges, such as: small data, unbalanced labels, large-ranging object scales, multi-class labels, and multimodal imaging, etc. To address these problems, in this project, as part of the MSD challenge, we propose a generic machine learning algorithm which we applied on two organs: liver and tumors, spleen. We propose an unsupervised generic model by implementing U-net CNN architecture with Generalized Dice Coefficient as loss function and also as a metric. The MSD dataset consists of dozens of medical examinations in 3D (per organ), we’ll transform the 3-dimensional data into 2-d cuts as an input of our U-net. Experimental results show that our generic model based on U-net and Generalized Dice Coefficient algorithm leads to high segmentation accuracy for each organ (liver and tumors, spleen), separately, without human interaction, with a relatively short run time compared to traditional segmentation methods.

ActiveSegmentation: A Simulation Framework for Benchmarking Active Learning Strategies for 3D Medical Image Segmentation

Sci. Rep. 2025 | Revisiting model scaling with a U-net benchmark for 3D medical image segmentation

MedSegMNIST is a python library for easy access to 2D and 3D Medical Image Segmentation datasets provided in varying resolutions/sizes, for benchmarking and learning purposes.

Welcome, Awesome-segmentation-in-medical project. This repo is robust, reproducible, and fair benchmarking framework for breast ultrasound (BUS) image segmentation. Features 10+ models (UNet, UNet++, TransUnet, SwinUnet etc.), 5+ datasets, and strict data separation to prevent test set leakage.

M3DA: Benchmark for Unsupervised Domain Adaptation in 3D Medical Image Segmentation

A Comprehensive Survey of Medical Image Segmentation: Challenges, Benchmarks, and Beyond

A Comprehensive Benchmark for Federated Learning on Medical Image Segmentation

M3DA: Benchmark for Domain Adaptation in 3D Medical Image Segmentation

A comprehensive and out-of-the-box benchmark for 3D medical image segmentation

The last benchmark library you will need for common medical image segmentation tasks

Privacy-aware medical image segmentation using conditional diffusion models (DDPM) and federated VMUNet. Accepted at (ICICS 2025); A unified benchmarking framework for image segmentation with UNet variants and Transformer-based models on ISIC 2018.; 基於基礎入門科研的影像分割

Semi-Supervised-Medical-Image-Segmentation-Benchmark

A Benchmark for 3D Medical Image Segmentation

This repository is heavily based on Medical-Image-Segmentation-Benchmarks

The Knee2024 benchmark dataset for medical image segmentation

Benchmarking Vison Models for 2D Medical Image semantic segmentation

Source code of our continual medical image segmentation benchmark paper.

Application of some segmentation approaches to medical images for benchmarking

Benchmarking Semi-Supervised Learning Approaches to Increase Medical Image Segmentation Performance

MedSegBench: A Comprehensive Benchmark for Medical Image Segmentation in Diverse Data Modalities

A comprehensive research-grade framework for benchmarking state-of-the-art medical image segmentation models on the ADAM (Aneurysm Detection And segMentation) dataset.

A repository of tensorflow based semantic image segmentation networks. The project not only provides various models but also includes data-loaders for both generic and medical benchmark datasets

SSGNet is a framework for data-efficient medical imaging that integrates class-specific StyleGAN3 generation with iterative semi-supervised pseudo-labeling. It enlarges datasets, alleviates class imbalance, and improves classification and segmentation performance across multiple benchmarks under limited annotations.

🧠 MedSegBench-SegLab is your all-in-one 🚀 plug-and-play lab for benchmarking medical image segmentation 🧩. Each notebook follows a 13-step reproducible pipeline—from caching to calibration—so you just swap your model in Step 6️⃣. Get live 📊 Plotly curves, 🩺 overlays, and ⚙️ speed + VRAM metrics for fair, Colab-ready, seed-locked comparisons! ✨