Choquet integral based deep learning model for COVID-19 diagnosis using eXplainable AI for NG-IoT models

Abstract

The COVID-19 outbreak has caused a global threat to the world healthcare system. The virus has mutated into different variants and mutations which spread more rapidly, are more deadly and have the lesser effect of vaccines. The number of cases is way more than the number of cases doctors can handle. Widely used methods of RT-PCR are tedious and time-consuming, Instant methods like antigen are very less effective and give incorrect results. Hence Artificial Intelligence (AI)-based Computer-aided Diagnosis (CAD) methods that help doctors in the correct and efficient diagnosis are the need of the hour. In this study, we propose a Deep Learning based XAI model for aiding clinical interpretation on Chest X-ray (CXR) images. The XAI diagnostic model can be integrated into NG-IoT devices using CAD models by provide accurate explainable diagnostics. We used deep-learning models with integrated Choquet integral as an aggregation function for more precise classification. Integrating it with the XAI model Grad CAM ++ increased its explainability. In our proposed work we used four different deep learning models ResNet-50, Inception V3, Densenet-121 and DCNN with integrated Choquet Integral function which increased the accuracy by 2% for Densenet-121 and DCNN and 1% for ResNet-50 and Inception V3. Using Grad CAM++ on the images increases the acceptance by medical professionals by clearly depicting how the models took the decision. We can depict that our proposed model Densenet 121 integrated with ChI outperformed all the other methods.