Enhancing Chest Disease Detection Through The Synergy Of Deep Learning And Genetic Algorithms

Résumé: Chest X-ray (CXR) imaging plays a pivotal role in modern healthcare, serving as a prominent and widely employed diagnostic modality. Its application is instrumental in aiding radiologists to discern various critical pulmonary conditions, enabling visual scrutiny of CXR images to identify disease manifestation. However, the intricate nature of lung diseases, characterized by resemblant patterns and symptoms, poses signifcant diagnostic challenges that may potentially result in severe ramifcations if misinterpreted. Deep learning (DL) models have surfaced as auspicious prediction methodologies, exhibiting remarkable precision akin to human-level performance. In this work, we are addressing the problem of lung disease detection using CXR images by employing a custom Deep Convolutional Neural Network (DCNN) architecture, we employ the model to extract intricate feature representations and accurately classify potential diseases within the CXR images. We are also deploying one of the Python Genetic Algorithm (GA) library PyGAD modules which is the pygad.gacnn module used for training DCNNs using the GA. The targeted dataset is the COVID-19-Radiography database. We investigate the deployment of the GA approach on three different datase samples. The results demonstrated the effect of the utilization of the GA allows us to achieve optimal outcomes by identifying the optimal set of hyperparameters and selecting the fttest individuals from each generation. Alternatively, the integration of specialized library modules, similar to the one employed in our investigation, enables the construction and training of CNN using the GA, facilitating the attainment of accurate results. The evaluation showed that the classifcation accuracy of our genetic algorithm approach achieved an Accuracy (ACC) of 77.31%, 78.23%, and 78.87%, improving the performances by 0.51%, 1.56%, and 1.7% for the three proposed samples respectively.

Publié dans la revue: