Related Experiment Videos
Diagnosis of breast cancer by measuring nuclear disorder using planar graphs
C Herrera-Espiñeira1, C Marcos-Muñoz, J E López-Cuervo
1Department of Pathologic Anatomy, Virgen de las Nieves General Hospital, Granada, Spain.
Objective:
To achieve a classifier of breast lesions to distinguish benign from malignant mammary lesions by quantifying nuclear disorder in epithelial cell groups from smears obtained by fine needle aspiration.
Study Design:
The study included 95 cases of breast cancer (289 groups) and 47 of benign breast lesions (150 groups), diagnosed by cytology. Information from planar graphs (mean of nuclear distances, standard deviation, maximum and minimum distance between nuclei) was used, and an algorithm constructed for this purpose was applied. The data were classified by double methodology--discriminant analysis, and classification and regression trees (CART)--to determine which achieved the best results.
Results:
CART selected the standard deviation of nuclear distances with accurate classification in 95.7% of benign lesions and 97.9% of malignant. Discriminant analysis constructed the discriminant function using the mean of nuclear distances and its standard deviation, with results similar to those of CART.
Conclusion:
The classifier based on nuclear disorder that we constructed proved to be rapid, simple and effective for malignant-benign discrimination in breast lesions and should be of diagnostic assistance.