A CNN-Based Solution for Breast Cancer Detection With Blood Analysis Data: Numeric to Image

Breast cancer has a high mortality rate worldwide. Early diagnosis is of great importance to reduce this rate. Early diagnosis ensures that patients receive appropriate treatment as soon as possible. Early diagnosis can be achieved by automatic classification through computerized systems. This study aims to detect breast cancer using frequently used blood analysis data recently. Data from the UCI library contains anthropometric values that can be collected in routine blood analysis, i.e. the features are numerical data. For high classification success, these numerical data are converted into image data. Then, the number of these images is increased by different data augmentation techniques and classification is performed with popular Convolutional Neural Network (CNN) models (AlexNet, ResNet50, and DenseNet201). Classification accuracy of 95.33% provided by ResNet50 is superior to other studies using the same data. In addition, converting numerical blood analysis images into image data is a strategy applied for the first time.