Hybridization between deep learning algorithms and neutrosophic theory in medical image processing: A survey

Abstract

Deep learning can successfully extract data features based on dealing greatly with nonlinear problems. Deep learning has the highest performance in medical image analysis and diagnosis. Additionally, deep learning performance is affected by insufficient medical image data such as fuzziness or incompleteness. The neutrosophic approach can enhance deep learning
performance with its great dealing with inconsistency and ambiguity information in medical data. This survey investigates the various ways in which deep learning is enhanced with neutrosophic systems and provides an overview and concept on each other. The hybrid techniques are classified based on different medical image modalities in different medical image processing stages such as preprocessing, segmentation, classification, and clustering. Finally, future works are also explored. In this study the highest accuracy was achieved by hybridization between neutrosophic and LASTM to classify the cardio views. While the highest capability to precisely detect those with the disease (sensitivity) is achieved by integration between neutrosophic, convolution neural network and support vector machine. Best specificity was obtained by neutrosophic and LSTM.