Revealing Multiple Lung Condition Using Deep Neural Network Models from X-RAY Images

This paper is about the predicting the accuracy of different optimal lung disease prediction models. Basically, in this, multiple lung diseases are detected by training on different models of CNN and used deep learning algorithm as it works well with large number of datasets as various dimensions and features of images as well as textual data can be investigated. Lung infections/diseases can be detected by taking blood tests etc. but it is expensive. So, this paper aims towards comparison done to find the most accurate model which can predict using x-rays to reduce the cost and time of the conduct. Image classification models and segmentation techniques like VGG19, NASNet, DenseNet, MobileNetV2, ResNetV2, EfficientNetB0, InceptionV3,etc. can be utilized to detect lung infections and illnesses. The primary limitation of traditional methods is the semantic gap between the high-level semantic information that humans perceive and the low-level visual information that imaging technologies gather. The deep convolutional neural network was developed to eliminate the difficulty in maintaining and querying enormous datasets. After performing different CNN models, DenseNet121gives the best accuracy. The obtained accuracy of the model after 15
^th
epoch is like 100 % of training accuracy, 0.0011 is training loss, 98.7% of validation accuracy and 0.0462 is validation error. The obtained results show the performance of various models, and which one is good and can be used as a secondary opinion tool by medical experts.