Autoencoder-based Efficient Denoising for Medical Images

The cost of an accurate medical diagnosis is extremely high, particularly in developing nations. Pneumonia is a prevalent ailment that poses a considerable barrier to medical diagnosis because of the striking similarities between its forms and the scarcity of medical pictures for related diseases in recent times. Transfer learning offers a potential way to move information from general to particular activities. Regrettably, testing and applying many transfer learning models don't produce acceptable outcomes. This study reveals a deep-learning algorithm based denoising approach using autoencoder developed for diagnosing medical image. Leveraging Convolutional Neural Networks (CNNs), the proposed method effectively removes noise from medical images while preserving diagnostically relevant features. This particular work focuses on denoising a medical image and then using it in classification of pneumonia using autoencoder and transfer learning. Autoencoder based preprocessing with an improved loss function and Gaussian distribution is used. Following this, classification is performed with the use of Four-Layered Convolution Neural Network (FCNN) and transfer learning to predict COVID 19 in chest CT images. A comparative study is also summarized with and without denoising process .Through extensive experimentation and evaluation of diverse medical imaging datasets, our approach reveals improved denoising performance when it is compared with traditional methods. Furthermore, the trained autoencoder shows robustness to varying levels of noise, ensuring consistent diagnostic accuracy of 99.52%. The developed framework offers a promising tool for enhancing the reliability of medical image interpretation, facilitating more accurate diagnosis and treatment planning in clinical settings.