Detecting lung diseases using deep learning algorithms

Abstract

Deep learning and computer vision are among the most important sciences of artificial
intelligence that have developed greatly and have increased interest in them in recent times, as this
development has been employed to build smart applications in various educational, industrial, commercial,
security, medical and other fields, and since medical data depends on images very much, there are uses A
wide range of computer vision in the medical field, the most important of which is modern diagnostic
methods and analysis of medical images. This research suggests the use of effective deep learning algorithms
to detect lung diseases, and because of the large number of them, only (covid-19 - Canser) was detected, as
the faster and accurate the detection of these two diseases, the greater the chances of successful treatment,
and thus the lower death rate for these two diseases. A deep learning model based on CNNs was proposed
to detect Covid-19 and cancer, and this model was trained on two databases, the first consisting of X-ray
images consisting of 6000 images divided into three sections 2000 healthy, 2000 infected with cancer and
2000 infected with Covid -19). The second consists of CT images consisting of 6000 images divided into
three sections (2000 healthy, 2000 infected with cancer, and 2000 infected with Covid-19). First, image
processing operations were performed to configure it to become with acceptable specifications for use in
the deep learning model, then a comparison was made between several pre-trained models, where our model
achieved the best evaluation accuracy of the previously trained models, and the results gave 86% accuracy,
98% sensitivity, and 72% specificity when the database was applied. The first, while the results were 98%
correct, 99% sensitive, and 97% specific, when the second database was applied. Thus, the results show the
effectiveness of the proposed algorithm as an assistant model for physicians and radiologists as an image
reader and to validate their results.