An Application of Advanced Neutrosophic Numbers for Lung Disease Detection

Abstract

Lung disease is one of the diseases that can be cured when the disease is detected in its early stages before it get progressed. Early diagnosis can be promising to curable. But, the most people fail to detect their disease before it comes to chronic. It leads to an increases in the death rate all around the world. Hence, an efficient model is required for the early detection and classification of lung diseases and its types. Generally, the Chest X-ray (CXR) are used to predict the different types of lung diseases like lung cancer, pneumonia, COVID-19 etc. But, the CXR images consist of fuzzy and imprecision information that makes segmenta­tion and classification a difficult task. Thus, the generalizations of fuzzy sets are used utilized to reduce the fuzziness and uncertainty in images. Fuzzy methods fails to consider the spatial pixel data due to noise and artifacts which play a vital role in many real-time applications. To resolve this, Consecutive Neutrosophic Set (CNSS) approach with Deep Learning (DL) model (CNSDL) is developed for lung diseases. In this model, Bipolar Trapezoidal Double Refined Indeterminate Neutrosophic Set (BTDRINSS) and Gaussian based Bipolar Trapezoidal Double Refined Indeterminate Neutrosophic Set (GBTDRINSS) is developed and employed along with the original NSS for the prediction tasks.  The suggested model is categorized into three stages. Initially, the input CXRimages are converted to three NS domain with True  False  and Indeterminacy  set images individually which reduces the fuzziness and retain more significant information for feature extraction of the opacity to distinguish the types of lung diseases.The entropy in three NS domain is applied to evaluate the indeterminacy. Then, the two operations like mean and enhancement operations are used to reduce the set indeterminacy to enhance the image edges to improve accuracy. In addition,  means clustering method is utilized for the image segmentation tasks. Finally, the enhanced image features are given as input to the DL model (ResNet 50) to train and test for the type of lung diseases.The test experimental outcomes show that the proposed GBTDRINSS model provides the accuracy of 97.25% on lung diseases prediction compared to the other existing models.