Better Application of Bayesian Deep Learning to Diagnose Disease

Abstract

A significant, but challenging, phase in healthcare is detecting and predicting status for a complex human disease. By the year 2050, the global population over 60 years of age will be 2 billion, according to the World Health Organization. For most health conditions, age is the principal cause. For the analysis of multi-dimensional patient's data, machine learning provides a principled approach to develop and compare sophisticated, automated and objective algorithms. Massive public health problems and economic burdens have been generated by complex human diseases such as cancers, Alzheimer, Parkinson, Motor Neuron, cardiovascular, and respiratory diseases. Technically, deep learning concept is derived machine learning under AI (Artificial Intelligence) that follows the brain structure and function in processing data and decision making. Deep learning uses many layers of nonlinear processing units to work with complex data sets and can transform them to analyze the outcome. It has inspired a lot of interest in its use for medical imaging problems. In order to ensure patient safety by clinical diagnostics, Bayesian Deep Learning makes more precise and intelligent predictions.The goal of this research was to optimize and implement a better algorithm for deep learning. A research field for growth is to complement deep learning with Bayesian thought. The performance of the method proposed is dependent on the sensitivity, specificity, accuracy and precision of the algorithms. The current implementation supports showing that a mixture of deep learning algorithms can be found in the best model. Machine Learning tools such as Tensorflow 2.0, Scikit-Learn, Pandas, Matplotlib & numpy are used for execution with powerful Python in this proposed application. Bayesian Deep Learning's proposed approach has outperformed current methods of diagnosis & prediction and demonstrates high precision and great potential to develop clinical tools. With more than 98 percent accuracy, one prototype and 2 data sets are used for clinical diagnosis and predictions for cancer and diabetes.