Classification of Early Breast Cancer using Convolutional Neural Networks

Abstract

Breast cancer is a life-threatening disease that impacts millions of people worldwide, necessitating heightened awareness and effective early detection strategies. Current methods for determining the presence and type of breast cancer in a clinical breast exam conducted by healthcare providers may be inaccurate, imprecise, and prone to error. The goal of this research project is to offer a second opinion for physicians to predict whether or not a patient has early onset breast cancer through a machine learning model developed from an image dataset, to significantly reduce the error in breast cancer staging, making the process more efficient in healthcare. Early detection is crucial for successful treatments of breast cancer, as treatment options in late-stage diagnosis of breast cancer can have a much worse prognosis. In the US, 240,000 cases of breast cancer are diagnosed per year in both men and women with a mortality rate of approximately 42,000 per year. [1] To approach this problem, I have tested an MLP Classifier, Logistic Regression, Ridge Classifier, Random Forest Classifier, Decision Tree Classifier, Support Vector Classifier, and trained a Convolutional Neural Network to compare various results and determine the best accuracy. The Convolutional Neural Network yielded impressive outcomes, reaching up to 79% testing accuracy and 98% training accuracy. The final model exhibits the capability to identify early signs of breast cancer and differentiate between malignant and benign tumours with relatively high accuracy showing the potential of AI in combination with medical imaging that can assist medical staff in diagnosis.