Accurate Detection Of COVID-19 Using Deep Learning Image Classifier,K-EfficientNet,and Chest X-Ray Images

Today we are witnessing a multi-sectoral revolution thanks to artificial intelligence.On a medical level,the application of this technology helps highquality patient diagnosis,the advice of a better therapy for a person based on his medical history,and the startling forecast of people’s health in the future based on its precision.This is not a complete list.Consider Natural Language Processing(NLP),which is used to label collections of medical data and the use of data from randomized trials to estimate the impact of treatment provided to specific patients.On the ILSVRC,Convolutional Neural Networks now surpass human capabilities.Deep learning approaches have become the de facto standard for a wide range of computer problems.They’re not just good at image processing and analysis;in fact,they’re outperforming existing techniques in areas like natural language processing,speech recognition,and so on.Our current research enables effective COVID-19 identification from chest Xrays at a level that exceeds that of professional radiologists.We bring three main contributions in our work:the design of the model architecture,the dataset collecting,and a new model training approach.To create an appropriate model optimized by Differential Neural Architecture Search,we first propose K-DNAS,an evolutionary method based on Neural Architecture search.The algorithm found the architecture EfficientNet as the best baseline and extended it to get the model named K-EfficientNet finally.We generate a big dataset called K-COVID by integrating six public datasets available online,which contains 14,124 X-Ray pictures of patients with Pneumonia or COVID-19 and patients with Normal X-Ray images.We next used the notion of modified progressive learning to train the model,which entails scaling the input CXR pictures from 112 × 112 to 448 × 448 while changing regularization parameters like Dropout,RandAugment,and Mixup.Small image sizes are trained using weak regularization,while larger images are trained using more robust regularization.When followed by the application of transfer learning on the ImageNet dataset and data augmentation,this global system allows us to achieve 97.3%accuracy and 100%sensitivity,and 100%Positive Predictive Value on COVID-19 detection,breaking the record on similar tasks.