Bimodal Profile Diagrams for Breast Cancer Classification Using Convolution Neural Network

This study introduces a bimodal sensing system designed to classify tumorous and nontumorous breast cancer using tactile and multispectral sensors. The proposed approach simplifies the diagnostic process by enabling the identification of two major types of breast cancer with a single system. This method converts multiple raw images into a representative profile diagram by capturing the most relevant mechanical and spectral properties, enhancing classification accuracy and improving computational efficiency. The separate convolutional neural network (CNN) models are utilized to identify tactile properties, such as depth, size, and stiffness, and multispectral characteristics, such as asymmetry, texture, and inflammation. These detected properties are then used to calculate tactile and multispectral indices, informed by domain knowledge, for cancer detection. These two indices classified malignant breast tumors with 83% accuracy and inflammatory breast cancer (IBC) with 81% accuracy, respectively. This method highlights the promise of a noninvasive, affordable diagnostic tool that can be utilized in routine clinical settings, particularly those lacking specialized radiological resources, to support early and accurate breast cancer diagnosis.