A sliding window framework for classifying high-resolution whole-slide microscopy and histopathology images using deep neural networks.
DeepSlide is an open-source deep learning framework for classifying high-resolution whole-slide images (WSIs) in digital pathology and microscopy. It uses a sliding window technique to divide large images into manageable patches, trains convolutional neural networks to recognize histologic patterns, and aggregates predictions for slide-level diagnosis. The framework was developed to achieve pathologist-level accuracy in tasks like lung adenocarcinoma subtyping.
Researchers and computational pathologists working with whole-slide images, particularly those in medical imaging, histopathology, and digital pathology who need to automate slide classification.
DeepSlide provides a complete, reproducible pipeline tailored for histopathology images, including preprocessing, model training, and visualization. It is open-source, supports public datasets, and is designed to bridge deep learning with clinical pathology workflows.
Code for the Nature Scientific Reports paper "Pathologist-level classification of histologic patterns on resected lung adenocarcinoma slides with deep neural networks." A sliding window framework for classification of high resolution whole-slide images, often microscopy or histopathology images.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Provides a complete workflow from data splitting to visualization in seven steps, ensuring reproducibility for computational pathology research as outlined in the README.
Automatically filters whitespace in H&E-stained images and balances class distributions during data processing, tailored for medical imaging tasks.
Includes a public lung adenocarcinoma dataset of 143 slides with pathologist annotations, facilitating easy benchmarking and validation without external data hunting.
Generates overlay images showing patch-level predictions on original slides, aiding in model debugging and clinical interpretation, as demonstrated in the visualization step.
Supports flexible hyperparameter tuning via argparse flags and architectures like ResNet-18, adaptable to various histopathology datasets.
Only supports single GPU training, which can slow down experiments and limit scalability for larger models or datasets, as noted in the known issues.
Primarily designed for Ubuntu with no guaranteed support for Windows, requiring additional setup efforts and potential bugs on other operating systems.
The patch generation process consumes significant disk space, as highlighted in data processing, which can be prohibitive for large studies without on-the-fly generation.
Requires multiple external libraries like OpenSlide and specific Python versions, making installation and maintenance non-trivial, especially for novice users.