Deep learning-based pneumonia classification using CNN models

• Main Authors: Saranya, S. S., Singh, Kashyapi, Agrahari, Aradhya, Prasanth, K.
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Algorithms; Bacteria; Chest; Datasets; Deep learning; Diagnosis; Effectiveness; Image classification; Image enhancement; Machine learning; Medical imaging; Pneumonia; X-rays
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0217211

One or both human lungs can become infected with pneumonia, a fatal bacterial condition brought on by pneumonia bacteria. Since the pandemic, interstitial lung disease has surpassed all other causes of death. The interpretation of chest X-rays used to diagnose pneumonia must be performed by qualified radiotherapists. Examining chest X-rays is difficult for a qualified radiologist, however. An early diagnosis of paediatric pneumonia might aid in hastening the healing process, evaluating, and enhancing diagnosis accuracy is vital. Therefore, developing an automated approach for identifying pneumonia may aid in hastening treatment, mostly in inaccessible regions. The deep learning algorithms’ effectiveness in examining radiology and the benefits of pre-trained CNN models’ features from large datasets in image classification applications allowed CNNs to receive much interest for illness categorization. With this research, one can examine how well pre-trained CNN models (VGG, ResNet, and Inception) perform. We propose a model for pneumonia diagnosis built utilising digitised chest X-ray images that may aid radiologists in their decision making. After comparison, VGG16 outperformed ResNet50 (84%) and InceptionV3 (65%) regarding classification accuracy, with a 90% score. These results suggest that for the particular dataset and task being studied, VGG16 was the most effective model.