Cell-Fie: A Machine Learning Tool to Improve the Accuracy of Fine -Needle Aspiration Biopsies in Diagnosing Breast Cancer

Overview

Breast cancer remains a leading cause of female mortality, claiming over 42,000 lives annually. Breast cancer is typically diagnosed using core -needle biopsies with very high accuracies; however, these procedures are very invasive, painful, and costly. An alternative, minimally invasive, and more affordable diagnosis method is called a Fine-Needle Aspiration Biopsy (FNAB), however, its accuracy is only 77.5%. This study aimed to improve FNAB breast cancer diagnoses by i) determining the significance of differences in 9 cell features of breast masses between cancerous and non -cancerous patients, and ii) use this information to identify the optimal machine learning model for breast cancer diagnosis based on these cell features. Analyzing data from 683 patients, differences in cell feature levels were evaluated using violin plots and Welch’s t -tests. Then, 5 machine learning models (logistic regression, decision tree, random forest, SVM linear, and SVM polynomial) were trained and compared using diagnosis accuracy, ROC curves, and AUC. The results showed that the cell features of clump thickness, uniformity of cell size, uniformity of cell shape, marginal adhesion, single epithelial cell size, bare nuclei, bland chromatin, normal nucleoli, and mitosis levels are significantly higher in breast cancer patients. When these cell features were used to predict breast cancer using machine learning, the random forest model had the highest diagnosis accuracy (99.26%) and AUC (99.8%). This random forest model can be a key tool in diagnosing breast cancer earlier, less invasively, cheaply, and very accurately, potentially leading to earlier treatment and improved breast cancer survival.

Competition history

• JSHS 2024 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.