A Noninvasive Approach Using Multi-tier Deep Learning Classifier for the Detection and Classification of Breast Neoplasm Based on the Staging of Tumor Growth

It is estimated on statistics that one out of eight females is affected by breast cancer worldwide. The scope of this work is to formulate a clinical protocol to find early-stage breast cancer and staging non-invasively. This work is to differentiate benign and malignant tumors with a novel ensemble approach. The ensemble utilization of signal processing by a recurrent neural network (RNN) and image processing by deep convolution neural network (DCNN) for the characterization of breast cancer yields the best result in disease prognosis. DCNN based system model named AlexNet is used to effectively classify breast neoplasm with optimum results in the mammogram dataset. In DCNN, the last layer called fully connected (FC) layer is linked to the support vector machine (SVM) classifier. The classifier fusion technique is adopted to combine the result of both imaging and signal processing to obtain the best classification result. Once it classifies with a greater value of true positivity it continues to go with Raman spectroscopy for the identification of spectral features associated with cells and tissues such as DNA, carbohydrate, nucleic acid, lipids, and proteins during the formation of breast neoplasm in the suspected candidate's blood plasma samples. Long Short-Term Memory (LSTM) based RNN is utilized to classify breast neoplasm features from the spectral dataset. After the classification, a combination of principal component analysis with factorial discriminant analysis (PCA-FDA) is used to find the different stages of cancer growth. This method promisingly dealt to have better specificity and sensitivity for all stages.