Silicosis Detection Using Extended Transfer Learning Model

Silicosis is a prevalent and damaging occupational disease in India. The detection of silicosis on X-ray images heavily relies on the expertise of radiologists, resulting in delayed detection and diagnosis. To overcome this challenge, machine learning-based computer-aided detection methods are being explored. However, due to the lack of publicly available large databases of silicosis images, highly accurate deep-learning models for silicosis detection through X-rays are difficult to obtain. In this study, we propose a method for silicosis detection using transfer learning techniques on X-ray radiographs. We trained and tested well-known deep transfer learning models, including Vgg16, Vgg19, ResNet50, ResNet101, Densenet121, Densenet169, Densenet201, Xception, Mobilnet, InceptionV3, and Efficient B7 models. Our proposed model is based on ResNet50 in which we add additional layers, which achieved an F1 score of 0.8166. Additionally, we applied the Grad-CAM technique to produce heatmap images that highlight the important features extracted from the X-ray images This approach could serve as a feasible alternative in countries with limited access to specialized radiological equipment and trained radiologists for detecting silicosis on X-ray images.