Non-Invasive Technique of Breast Cancer Diagnosis Using Interpretation of Fractal Dimension of Cells Nuclei in Buccal Epithelium
Dmitriy A. Klyushin, O. Kravets, Kateryna Golubeva, N. Boroday · 2023
The paper describes novel machine-learning, high-precision methods for diagnosing breast cancer based on the fractal dimension of buccal epithelium nuclei. The method uses the Random Forest algorithm and logistic regression applied to data on fractal dimension of images estimated by blue, green, and red channels. We investigated the control group (29 woman), patients with breast cancer at the second stage (68 woman), and patients with fibroadenomatosis (33 woman). The dataset consists of 20256 photos of interphase nuclei of buccal epithelium (6752 nuclei photographed without filter, through a yellow filter, and through a violet filter). All images were preprocessed using tools of morphological analysis. The best results are obtained when using all data channels: 95% accuracy, 92% sensitivity, and 100% specificity. The fractal dimension is estimated using the Hurst exponent. High performance and non-invasiveness of the method gives the hope that it well be useful for clinical applications.