Using Deep Learning and Particle Swarm Optimization Algorithm to Recognize Fatty Liver in Compressed Images

Abstract

With the increasing rates of liver steatosis and fibrosis, the need for accurate and rapid diagnostic tools for these conditions using medical imaging has become pressing. This study aims to develop a deep learning-based model for analyzing and diagnosing liver steatosis from compressed medical images, with performance enhancement using the Particle Swarm Optimization (PSO) algorithm. To improve the efficiency of the analysis process and reduce diagnostic costs, the Singular Value Decomposition (SVD) technique was employed for image compression, contributing to data size reduction while preserving the essential image quality. The proposed model relies on an Autoencoder Network supported by the PSO algorithm to enhance image quality after compression and was implemented using MATLAB software. Previous studies demonstrated the effectiveness of PSO in improving analysis accuracy, achieving 92.2% accuracy, an F-Score of 87.2%, and an Intersection over Union (IoU) of 90.7%. In the current study, the results showed that the developed model achieved higher performance, with an accuracy of 94.81%, an F-Score of 97.34%, and a Jaccard Index of 94.81%. Additionally, the model successfully reduced image sizes by 45% with a low execution time of approximately 6.27 seconds, without compromising the quality of the medical analysis. This model represents a promising tool in non-invasive medical imaging for studying liver diseases, enhancing the speed and accuracy of diagnosis, assisting physicians in making informed clinical decisions, and reducing the costs and risks associated with traditional diagnostic methods.