AN EFFICIENT BREAST CANCER PREDICTION SYSTEM USING A MODIFIED DEEP NEURAL NETWORK (DNN)

Abstract

A quantum inspired modification of the Deep Neural Network (DNN), for breast cancer prediction is presented; it exhibits improved convergence and classification accuracy. With softmax output activation, it maps patient feature vectors to probability distributions over cancer subtypes in a multi layer architecture. The backpropagation is used to minimize the categorical cross entropy loss function, thus calculating the weight and bias gradient updates. The weight update rule is transformed by a unitary matrix in such a way that they are better stable and robust, applying a quantum inspired transformation. Quantum regularization is incorporated in the gradient update equation so as to eliminate the vanishing gradients and have better generalization. The proposed pseudocode outlines the forward propagation, loss computation, backpropagation, and novel weight update process. The Quantum Computing based Deep Neural Network (Q-DNN) is found to be better than traditional DNN for classification accuracy, convergence speed and resilience to local minima, which makes it a potential progress towards breast cancer detection with deep learning.

KEYWORDS:

Breast cancer, medical diagnosis, quantum computing, deep learning, accuracy, and multilabel classification.