MIScnn: a framework for medical image segmentation with convolutional neural networks and deep learning.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Modern medical imaging advancements necessitate automated segmentation solutions.
• Existing platforms lack user-friendly tools for setting up medical image segmentation pipelines.
• Current pipelines are often standalone and dataset-specific.

Purpose of the Study:

• Introduce MIScnn, an open-source Python library for medical image segmentation.
• Provide an intuitive API for rapid pipeline construction, including data handling, preprocessing, and model training.
• Enable high configurability and customization for diverse segmentation tasks.

Main Methods:

• Developed MIScnn with an intuitive API for building segmentation pipelines.
• Integrated functionalities for data I/O, preprocessing, augmentation, and patch-wise analysis.
• Included a library of deep learning models and tools for training, prediction, and evaluation.

Main Results:

• Successfully ran a cross-validation on the Kidney Tumor Segmentation Challenge 2019 dataset.
• Developed a powerful predictor using a standard 3D U-Net model for multi-class semantic segmentation.
• Demonstrated rapid pipeline setup with minimal code.

Conclusions:

• MIScnn enables researchers to quickly establish complete medical image segmentation pipelines.
• The framework facilitates rapid development and customization of medical image analysis tools.
• Source code is publicly available for community contribution and utilization.