An Efficient PM - Multisampling Image Filtering with Enhanced CNN Architecture for Pneumonia Classfication

• Published in: Biomedical signal processing and control 2023-09, Vol.86, p.105296, Article 105296
• Main Authors: Nithya, T.M., Rajesh Kanna, P., Vanithamani, S., Santhi, P.
• Format: Article
• Language: English
• Subjects: Convolutional Neural Networks (CNNs); Deep Convolutional Generative Adversarial Network (DCGAN); Generative Adversarial Networks (GANs); Perona-Malik (PM)
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2023.105296

•The proposed PM-Multisampling filtering technique preserves the edges during the smoothening process of the images.•The PM-Multisampling filtering technique takes the multiple parts of the images for smoothening by adopting the block averaging and Gaussian filtering techniques.•The class imbalance problems are addressed with Multipiled DCGAN by generating the synthesized image usingmultistacked CNN.•The performance of the augmentation and classification is combined to reduce the time taken to complete the process, accurate prediction without loss and free from overfitting and underfitting. Pneumonia detection using deep convolutional networks is an important application of deep learning in the field of medical image analysis. In this approach, a Convolutional Neural Network (CNN) is used to classify chest X-ray images as either normal or indicating the presence of pneumonia. The CNN is trained on a dataset of labeled images, learning to recognize patterns associated with pneumonia, such as infiltrates or consolidation in the lungs. The following three problems are addressed in this research work. Initially, the images with noise are not suitable for prediction and most of the filtering techniques generate the loss in the data and over smoothening of images. It can be resolved by the proposed Perona-Malik (PM)- Multisampling technique. It preserves the edges and smoothens the different region of image parallel manner. Second, the class imbalance problem is addressed by using synthesized image generation using Generative Adversarial Networks (GANs). So, the equal number of train and test images considered for better accuracy. Third, the accurate prediction is enhanced by Multi piled Deep Convolutional Generative Adversarial Network (DCGAN). It produces the accuracy of the model as 96% and loss as 10% without overfitting and under fitting. The execution time of the proposed model is 7.2 s from the initial stage to final stage.