Toward an Efficient Deep Learning Model for Lung pathologies Detection In X-ray Images
Abdelbaki Souid, Nizar Sakli, Hédi Sakli
Abstract
Medical imaging methods identify and record anomalies in the human body. These techniques are critical for assessing, diagnosing, and treating lung illnesses. Chest radiography (chest X-ray) is a low-cost yet effective medical imaging technology. However, a scarcity of trained radiologists may significantly restrict the technique's use. To detect abnormality in chest radiographs, emerging technologies such as deep learning should be applied to improve diagnostic performance and accuracy. CNN-based deep learning algorithms have made significant progress. The success of CNN in image classification has led researchers to investigate its utility as a diagnostic method for identifying and characterizing lung diseases. To achieve the defined goal, we leverage and extend the EfficientNet family of deep artificial neural networks knowns for their high accuracy and small footprint in other applications. A collection of three datasets is used to train the proposed approach. The results show that the proposed approach was able to produce a high-quality model, with an overall AUC of 0.871, and the overall sensitivity of 79.4%, while having 5 to 30 times fewer parameters than other architectures.