Lung Diseases Identification from Chest Radiographs based on Transfer Learning

In the year 2019 global health crisis rapidly evolved into a worldwide medical emergency. This respiratory illness presents signs like pneumonia, tiredness, harsh cough, as well as high temperature that may confused you with other conditions like lung cancer, tuberculosis (TB), or typical pneumonia. As a result, rapid and accurate diagnosis is crucial to reduce mortality rates among patients. Medical facilities frequently use chest X-rays (CXR) for quick and precise diagnoses. Deep learning (DL) algorithms have shown remarkable potential in identifying various kinds of diseases that affect human lungs, significantly expediting the process of diagnosing various lungs disease supporting medical professionals. The dataset consists of CXR images: 3616 images of coronavirus disease, 6012 images of lung opacity disease, 1345 images of Pneumonia contaminated lungs, also 10,192 images of lungs that are normal. These images are pre-processed, which include resizing of the images also the normalizing and at random divided them in order to fulfil the requirements of our neural network model based deep learning architecture. The implementation utilizes transfer learning, employing a previously trained ResNet50 architecture, afterward additional convolutional units and fully connected layers. The experimental results demonstrated that the ResNet50 + Fully Connected NN model surpassed existing methods, attaining 97.09% of overall accuracy, 96.97% of recall value, 97.17% of precision value, 92.30% of F1 score, and an 98.85% of AUC value (Area Under Curve). This model offers an efficient tool for physicians to examine also manage lung diseases further effectively, faster patient treatment and care.