Related Experiment Video
Updated: May 7, 2026
08:16
X-ray Visualization of Intraductal Ethanol-based Ablative Infusion for Prevention of Breast Cancer in Rabbit Models
Published on: September 12, 2025
Breast cancer characterization based on image classification of tissue sections visualized under low magnification
C Loukas1, S Kostopoulos, A Tanoglidi
1Department of Medical Physics, Medical School, University of Athens, 75 Mikras Asias Street, 115 27 Athens, Greece.
Computational and Mathematical Methods in Medicine
|September 27, 2013
Summary
This study developed an automated system for breast cancer diagnosis using texture analysis and pattern recognition algorithms. The system achieved high accuracy in classifying malignancy grades, offering a faster alternative to manual histopathological examination.
Area of Science:
- Histopathology
- Computational Pathology
- Medical Imaging Analysis
Background:
- Accurate and rapid assessment of breast cancer tissue biopsies is crucial for diagnosis.
- Traditional histopathological evaluation relies on subjective assessment of nuclear shape and tissue architecture at different magnifications.
- Automated methods are needed to improve efficiency and consistency in breast cancer diagnosis.
Purpose of the Study:
- To develop and validate a pattern classification system for automated assessment of breast cancer images.
- To evaluate the performance of different pattern recognition algorithms using textural features from low-magnification images.
- To explore the potential of automated analysis for improving the speed and objectivity of breast cancer diagnosis.
Main Methods:
- Texture analysis was performed on 65 regions of interest from 60 breast cancer tissue images (×10 magnification).
- Thirty textural features were extracted per image.
- Three pattern recognition algorithms (kNN, SVM, PNN) were employed for classifying images into three malignancy grades (I-III).
- Classifiers were validated using leave-one-out (training) and cross-validation (testing) methods.
Main Results:
- The kNN and PNN classifiers achieved approximately 97% discrimination efficiency in training mode, while SVM achieved 95%.
- In testing mode, the PNN classifier demonstrated the highest classification accuracy at 90%, followed by kNN (86%) and SVM (85%).
- The proposed technique shows potential for analyzing complex, large-scale histopathological images.
Conclusions:
- Automated assessment of breast cancer tissue sections using textural features and pattern classifiers is feasible.
- The developed system offers benefits of speed and automation, potentially complementing or replacing manual visual examination.
- This approach can aid in the rapid and objective evaluation of breast cancer malignancy grades.
