TIAToolbox as an end-to-end library for advanced tissue image analytics

Area of Science:

• Computational pathology
• Digital pathology
• Bioinformatics

Background:

• Computational pathology is rapidly advancing due to deep learning.
• Whole-slide images (WSIs) present challenges in size and complexity for analysis.
• Lack of open-source, end-to-end APIs hinders widespread adoption of advanced algorithms.

Purpose of the Study:

• To present TIAToolbox, a Python toolbox for computational pathology.
• To make computational pathology accessible to a broader range of researchers.
• To provide a generic, end-to-end API for pathology image analysis.

Main Methods:

• Developed modular and configurable components for common pathology image analysis tasks.
• Included functionalities for reading WSIs, patch extraction, stain normalization, augmentation, model inference, and visualization.
• Provided a user-friendly API for commonly used methods and models.

Main Results:

• Demonstrated the construction of a complete computational pathology deep-learning pipeline using TIAToolbox.
• Showcased how state-of-the-art deep-learning algorithms can be reimplemented with minimal effort.
• Validated the library's capability to streamline the development process.

Conclusions:

• TIAToolbox offers a usable and adaptable library for computational pathology.
• The toolbox provides efficient, cutting-edge, and unit-tested tools for the entire analysis workflow.
• Enables researchers to leverage recent deep-learning advancements in pathology image analysis.