3D whole body preclinical micro-CT database of subcutaneous tumors in mice with annotations from 3 annotators

Area of Science:

• Preclinical imaging
• Medical image analysis
• Oncology research

Background:

• Micro-Computed Tomography (µCT) is crucial for evaluating anticancer drug efficacy in mouse models with subcutaneous tumors.
• Manual tumor delineation in µCT scans is time-consuming and lacks consistency, necessitating automated segmentation (AS) tools.

Purpose of the Study:

• To introduce a comprehensive preclinical µCT database for developing and validating automatic segmentation (AS) tools for subcutaneous tumors.
• To establish a baseline for human annotation accuracy in tumor delineation.

Main Methods:

• Creation of the largest preclinical µCT database to date, containing 452 whole-body scans from 223 mice with subcutaneous tumors.
• Inclusion of ten diverse µCT datasets acquired between 2014 and 2020 using a preclinical PET/CT scanner.
• Manual annotation of each tumor by three expert annotators to enable robust model development and inter-annotator agreement analysis.

Main Results:

• The database comprises 452 whole-body µCT scans from 223 mice, representing the largest dataset of its kind.
• Analysis of inter-annotator agreement yielded a Fleiss' Kappa of 0.903 ± 0.046 and a mean volume estimation deviation of 0.015 ± 0.010 cm³ (6.9% ± 4.7%).
• These results establish a human baseline accuracy for subcutaneous tumor delineation and demonstrate good inter-annotator consistency.

Conclusions:

• The developed preclinical µCT database is essential for advancing automatic segmentation (AS) tools in oncology research.
• The established human annotation baseline provides a benchmark for evaluating the performance of AS algorithms.
• This resource will accelerate the development of more efficient and reliable methods for analyzing anticancer molecule efficacy in preclinical studies.