Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Yucca: A Deep Learning Framework For Medical Image Analysis

Sebastian Nørgaard Llambias, Julia Machnio, Asbjørn Munk, Jakob Ambsdorf, Mads Nielsen, Mostafa Mehdipour Ghazi

TL;DR

Yucca targets the gap between highly automated end-to-end medical-imaging frameworks and flexible, modular platforms. Built on PyTorch and PyTorch Lightning, it implements a three-tier architecture consisting of Functional, Modules, and Pipeline to support stateless building blocks, OO-wrapped components, and end-to-end workflows, respectively. The pipeline covers task conversion, preprocessing with planners, training via a Manager and LightningModule, and inference/evaluation with consistent post-processing and comprehensive metrics, achieving state-of-the-art results on diverse tasks such as cerebral microbleeds, white matter hyperintensity, hippocampus, and brain lesions. The framework aims to reduce engineering overhead, improve reproducibility, and invite community contributions, offering both novice-friendly out-of-the-box performance and advanced customization for researchers.

Abstract

Medical image analysis using deep learning frameworks has advanced healthcare by automating complex tasks, but many existing frameworks lack flexibility, modularity, and user-friendliness. To address these challenges, we introduce Yucca, an open-source AI framework available at https://github.com/Sllambias/yucca, designed specifically for medical imaging applications and built on PyTorch and PyTorch Lightning. Yucca features a three-tiered architecture: Functional, Modules, and Pipeline, providing a comprehensive and customizable solution. Evaluated across diverse tasks such as cerebral microbleeds detection, white matter hyperintensity segmentation, and hippocampus segmentation, Yucca achieves state-of-the-art results, demonstrating its robustness and versatility. Yucca offers a powerful, flexible, and user-friendly platform for medical image analysis, inviting community contributions to advance its capabilities and impact.

Yucca: A Deep Learning Framework For Medical Image Analysis

TL;DR

Yucca targets the gap between highly automated end-to-end medical-imaging frameworks and flexible, modular platforms. Built on PyTorch and PyTorch Lightning, it implements a three-tier architecture consisting of Functional, Modules, and Pipeline to support stateless building blocks, OO-wrapped components, and end-to-end workflows, respectively. The pipeline covers task conversion, preprocessing with planners, training via a Manager and LightningModule, and inference/evaluation with consistent post-processing and comprehensive metrics, achieving state-of-the-art results on diverse tasks such as cerebral microbleeds, white matter hyperintensity, hippocampus, and brain lesions. The framework aims to reduce engineering overhead, improve reproducibility, and invite community contributions, offering both novice-friendly out-of-the-box performance and advanced customization for researchers.

Abstract

Medical image analysis using deep learning frameworks has advanced healthcare by automating complex tasks, but many existing frameworks lack flexibility, modularity, and user-friendliness. To address these challenges, we introduce Yucca, an open-source AI framework available at https://github.com/Sllambias/yucca, designed specifically for medical imaging applications and built on PyTorch and PyTorch Lightning. Yucca features a three-tiered architecture: Functional, Modules, and Pipeline, providing a comprehensive and customizable solution. Evaluated across diverse tasks such as cerebral microbleeds detection, white matter hyperintensity segmentation, and hippocampus segmentation, Yucca achieves state-of-the-art results, demonstrating its robustness and versatility. Yucca offers a powerful, flexible, and user-friendly platform for medical image analysis, inviting community contributions to advance its capabilities and impact.
Paper Structure (11 sections, 3 figures)

This paper contains 11 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Framework Overview
  3. Functional
  4. Modules
  5. Pipeline
  6. Task Conversion
  7. Preprocessing
  8. Training
  9. Inference and Evaluation
  10. Results
  11. Conclusion

Figures (3)

  • Figure 1: The Yucca pipeline consists of four main modules: task conversion, preprocessing, model training, and inference and evaluation. Task conversion structures raw data and splits it into training-validation and testing sets. Preprocessing prepares the training data according to the selected YuccaPlanner. Model training uses the preprocessed data to train the selected model architecture, managed by the YuccaManager. Finally, the inference and evaluation preprocesses the test data, generates predictions, and computes performance metrics using the YuccaEvaluator.
  • Figure 2: Examples of augmentation techniques provided by Yucca. The first column shows the original T1 image, while the subsequent columns display the image with various augmentation techniques applied. The rows represent different anatomical views in (x, y, z) directions: sagittal (bottom), coronal (middle), and axial (top).
  • Figure 3: Comparison of the hippocampus segmentation results for a sample T1 MRI using different 2D-3D U-Net architectures.