Related Experiment Video
Updated: Jun 9, 2025
11:43
Advanced Animal Model of Colorectal Metastasis in Liver: Imaging Techniques and Properties of Metastatic Clones
Published on: November 30, 2016
12.6K
Classifying histopathological growth patterns for resected colorectal liver metastasis with a deep learning analysis
Diederik J Höppener1, Witali Aswolinskiy2, Zhen Qian1
1Department of Surgical Oncology and Gastrointestinal Surgery, Erasmus MC Cancer Institute, University Medical Center Rotterdam, Rotterdam, The Netherlands.
BJS Open
|October 29, 2024
Summary
A new deep-learning algorithm accurately classifies colorectal liver metastases growth patterns, aiding prognosis. This automated method shows promise for clinical use in pathology.
Area of Science:
- Oncology
- Digital Pathology
- Artificial Intelligence
Background:
- Histopathological growth patterns are key prognostic factors for colorectal liver metastases.
- Automated scoring methods are needed for efficient, objective assessment in practice and research.
- This study focuses on developing an automated method for classifying desmoplastic vs. non-desmoplastic growth patterns.
Purpose of the Study:
- To develop and validate a deep-learning algorithm (neural image compression) for classifying colorectal liver metastases growth patterns.
- To distinguish between desmoplastic and non-desmoplastic histopathological growth patterns using digital slides.
- To assess the correlation of automated classifications with patient survival.
Main Methods:
- Developed a deep-learning algorithm using whole-slide images from a Dutch single-center cohort (n=932).
- Externally validated the algorithm on whole-slide images from another Dutch institution (n=870).
- Classified dichotomous hepatic growth patterns and correlated with overall survival.
Main Results:
- The neural image compression algorithm achieved high discriminatory power (AUC 0.93-0.95) for classifying desmoplastic growth patterns.
- Automated and manual classifications showed similar prognostic value regarding overall survival.
- The algorithm demonstrated strong performance in both development and external validation cohorts.
Conclusions:
- The neural image compression approach is effective for pathology-based classification of colorectal liver metastases.
- Automated growth pattern assessment shows potential for clinical integration.
- This technology can enhance prognostic accuracy and patient management.
