Quantifying Nuclear Structures of Digital Pathology Images Across Cancers Using Transport-Based Morphometry.

Area of Science:

• Computational pathology
• Quantitative biology
• Cancer imaging analysis

Background:

• Nuclear morphology is crucial for cancer diagnosis and grading.
• Machine learning and large datasets offer new avenues for extracting insights from nuclear images.
• Existing methods may not fully capture the information content of nuclear structure.

Purpose of the Study:

• To introduce a novel transport-based morphometry (TBM) framework for modeling nuclear chromatin structure.
• To demonstrate the robustness and interpretability of the TBM framework across diverse datasets and cancer types.
• To establish TBM as a quantitative tool for comparative cancer studies.

Main Methods:

• Developed a TBM framework using optimal transport mathematics to model nuclear information content.
• Represented information content of each nucleus relative to a template nucleus.
• Applied the method to diverse cancer imaging data, including liver, thyroid, lung, and skin tumors.

Main Results:

• The TBM model effectively captures nuclear chromatin structure information.
• The framework is robust to variations in staining and imaging protocols.
• Demonstrated ability to differentiate benign from malignant nuclear features across multiple cancer types.

Conclusions:

• The TBM framework provides a quantitative approach to nuclear morphometry.
• This method enables meaningful comparisons across different datasets and cancer types.
• TBM has the potential to enhance cancer studies, technologies, and clinical applications.