Related Experiment Videos
Metastasis detection from whole slide images using local features and random forests
Mira Valkonen1,2, Kimmo Kartasalo1,2, Kaisa Liimatainen1,2
1BioMediTech and Faculty of Medicine and Life Sciences, University of Tampere, Tampere, Finland.
Summary
This study introduces a machine learning approach for automated detection of cancerous tissue in digital pathology slides. The method accurately identifies metastatic areas in lymph node samples, improving diagnostic efficiency.
Area of Science:
- Digital pathology
- Computational pathology
- Machine learning in histopathology
Background:
- Digital pathology necessitates automated methods for identifying cancerous tissues in whole slide images.
- Machine learning and image analysis offer potential for increased throughput and cost savings in histological assessment.
Purpose of the Study:
- To develop and evaluate a machine learning approach for detecting cancerous tissue in scanned whole slide images.
- To assess the accuracy and generalizability of the developed method for breast cancer metastasis detection.
Main Methods:
- Utilized feature engineering and supervised learning with a random forest model.
- Extracted local descriptors related to image texture, spatial structure, and nuclei distribution.
- Evaluated the method on lymph node samples for breast cancer metastasis detection.
Main Results:
- Achieved high accuracy in detecting metastatic areas (AUC 0.97-0.98 for tumor detection, 0.84-0.91 for tumor vs. normal tissue).
- Demonstrated good generalization across images from multiple laboratories.
- Generated an interpretable classification model linking features to tissue differences.
Conclusions:
- The developed machine learning method provides accurate and generalizable detection of cancerous tissue in digital pathology.
- The approach offers potential for enhancing efficiency and reducing costs in histological assessments.
- The interpretable model aids in understanding feature-based tissue differentiation.