Downloads: 1 | Views: 76 | Weekly Hits: ⮙1 | Monthly Hits: ⮙1
Research Paper | Computer Science and Information Technology | India | Volume 12 Issue 10, October 2023 | Rating: 5 / 10
Bat Algorithm with Variational Autoencoder for Pneumonia Detection and Classification
Parthasarathy V. | Saravanan S. [5]
Abstract: Pneumonia detection and classification are essential in earlier diagnosis and efficient treatment of this respiratory disease. Deep learning (DL) methods are developed as robust tools in medical imaging, providing the possibility to revolutionize the accuracy and effectiveness of this diagnostic method. This studyintroduces a Bat Algorithm with Variational Autoencoder for Pneumonia Detection and Classification (BAVAE-PDC) approach on Chest X-rays (CXR). To enhance the quality of CXR images, we exploit a median filter (MF) for preprocessing. The MF efficiently decreases noise and improves image clarity, offering a cleaner input for the following analysis. For feature extraction, we employ the state-of-the-art EfficientNet architecture. EfficientNet's deep neural model is extremely efficient at capturing complex patterns in medical images, permitting for the extraction of selective features vital for accurate classification. Classification is carried out employing a Variational Autoencoder (VAE). VAEs are called for their capability to model intricate data allocations and produce important latent representations. To improve the performance of the classification method, we utilize the Bat Algorithm (BA). The BA is a bio-inspired optimization algorithm that successfully tunes the hyperparameters of the VAE model, ensuring optimum effectiveness with respect to reliability and accuracy. The developed architecture is estimated on a significant database of CRX images, representing its efficiency in pneumonia detection and classification. The outcomes specify a remarkable enhancement in performance compared with conventional techniques.
Keywords: Chest X-Ray, Variational Autoencoder, Bat Algorithm, Medical Imaging, Deep Learning
Edition: Volume 12 Issue 10, October 2023,
Pages: 1844 - 1851