Breast Cancer Detection through Machine Learning Techniques

Abstract

The lifetime probability of developing Breast cancer for a female is 1 in 8. Given this probability, breast cancer becomes one of the most rapidly spreading and the most common type of cancer diagnosed in women. Although the cure is available in most countries and survival rate is increasing due to the advances in medical research, but still there is a huge gap in identifying this disease at initial stages. Pakistan, being a third world nation, alone, has the highest rate of increasing breast cancer in Asia with 90000 new cases every year out of which 40000 die. Various reasons contribute to this, lack of awareness among people, cultural setbacks, lack of research in medical, old treatments etc. Numerous cases go undiagnosed or reach last stage where treatments are ineffective, money is also wasted, and precious lives are lost. Traditional methods of diagnosis include breast self-examination, clinical examination, Biopsy, Mammography, CAD (Computer Aided Diagnosis), MRI and breast ultrasound. It is very difficult to identify a disease based on visual diagnosis of tissue including multiple features such as this one. In recent past, Machine learning techniques have been proven helpful to radiologists and pathologists for fast and efficient detection of breast cancer. Present investigation explores several machine learning and deep learning techniques to detect and classify breast cancer using transfer learning via VGG-19 which has not been previously done. Six machine learning techniques, Logistic regression, K nearest Neighbor (KNN), Support Vector Machine (SVM), Naïve Bayes , Decision tree and Random Forest whereas one deep learning technique , Artificial Neural Network (ANN) were applied. Three publicly available breast xi cancer image datasets were used in this research and the results were analyzed using various parameters i.e. Accuracy, Precision, Recall, F1 score and Specificity. It was observed from the results that Artificial neural network outperforms all techniques, detecting breast cancer with an average accuracy of 91% followed by Logistic Regression giving an average accuracy up to 90% for all three datasets. In conclusion, the results show the potential of accurate classification of breast cancer images as malignant or benign and proves to be useful in effective treatment of the disease as compared to traditional methods.