Early and precise diagnosis of pulmonary diseases is decisive for treatment so that spread of infection can be terminated. The provision of Computed Tomography is not available in remote places due to it's high cost and complexity. A publicly available dataset of more than 21,000 X-rays which contained normal, covid affected, viral pneumonia affected and lung opacity images has been used. A supervised neural framework is presented in this work that transfers acquired information and tunes performance in the associated task to categorise CXRs as viral pneumonia, covid 19, healthy lungs, and pulmonary opacity. The classification task is executed with seven state of art networks namely, DenseNet121, DenseNet 201, InceptionV3, ResNetS0V2, RetNet101V2, VGG19 and Xception to study the maximum accuracy that could be obtained for training. To measure the performance of the proposed technique, sensitivity, precision, confusion matrix, and Fl-score were employed in addition to accuracy. A classification accuracy of 89.4 percent was obtained on test set with ResNet101V2. The formulated model is then used to evaluate similarity index to calculate the percentage of pulmonary damage/opacity caused due to specific disease by comparing histograms obtained by ImgCompare function. The proposed model is not limited to the above mentioned classes and can also be used to determine other abnormalities if provided with corresponding data sets.
Deep Supervised Learning Based Severity Analysis on Chest X-ray
02.12.2021
1137745 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Traffic accident severity prediction based on interpretable deep learning model
Taylor & Francis Verlag | 2025
|Deep Self-Supervised Learning Models for Automotive Systems
British Library Conference Proceedings | 2021
|Using Deep Learning in Severity Analysis of At-Fault Motorcycle Rider Crashes
Transportation Research Record | 2018
|