Related Experiment Video For Breast cancer
Updated: Dec 22, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Computer-assisted Diagnosis of Breast Cancer by Cell Network Matrix Extraction and Multilayer Perceptron Analysis
Amir Momeni Boroujeni1, Elham Yousefi1, M A Haseeb1,2,3
1Department of Pathology, Kings County Hospital Center, Brooklyn, New York, USA.
Objective:
Diagnosis of breast cancer is based on identification of various morphologic features by histopathologic examination of the specimen. Ancillary immunohistochemical and molecular analyses provide additional information that is prognostic and guides therapy. Because of subjectivity in this approach, we sought to develop a computer model which could assist in differentiating normal or benign tissue from malignant breast carcinoma.
Methods:
Cases of benign sclerosing adenosis (20) and high-grade invasive ductal carcinomas (20) of breast were retrieved and re-examined. Five images of the diagnostic areas from each case were captured (400x). Each image was divided into quadrants and saved as 1-megapixel each. These 800 images were then binarized and segmented using the watershed method. The cell graphs were extracted to identify the matrix of adjacent cells and the network properties were determined for each image. The local network features were fed into a PAM model and global network features were fed into a multilayer perceptrons (MLP) to distinguish between benign and malignant samples. These characteristics were evaluated by training the models on 40% (320) of the randomly assigned images followed by real-time testing of the remaining 60% (480) images. In addition, normal breast tissue from five cases was retrieved and forty (40) images were captured to further test the model.
Results:
Both local and global network feature models had high area under the curve (AUC) (0.63 and 0.99 respectively), with their adjusted Rand indices (ARI) being 0.61 and 0.87, respectively. Pooling the pseudoprobabilities of the two neural networks greatly increased the accuracy of the model with predictions of the combined model at image level being 100% accurate with AUC of 1.
Conclusion:
This study shows that using a combination of cell-graph extraction and a deep learning algorithm computers can accurately distinguish between benign and malignant breast lesions.
