Classification of Gougerot-Sjögren Syndrome Based on Artificial Intelligence

Gougerot-Sjögren syndrome (GSS) is an incurable chronic autoimmune disease that involves an inflammatory process and lymphoproliferation that primarily affects the lacrimal and salivary glands. This disease mainly affects women (the ratio of affected women can be nine times higher than the ratio of affected men). According to an epidemiology study, GSS at different severity levels may affect between 0.1 and 5% of the total population. Usually, GSS detection is performed by biopsy. Some medical studies showed a correlation between biopsy results and the salivary gland ultrasonography (SGUS). On the other side, ultrasound imaging devices are widely used in various medical fields thanks to their noninvasive nature, safety and nonimpact on patients’ health. However, these grey images are affected by noise and artifacts. In our project, we developed an artificial intelligence approach to classify and detect GSS only based ultrasound imaging. Indeed, the salivary glands are made of tissue, with acinar, ductal, and myoepithelial cells. Some sonographic features are clearly identified for the detection of the primary GSS. Additionally, some patterns in the textures can help differentiate GSS with other diseases. So, we extracted specific features and then developed a learning scheme for deep neural networks based on joint training on classification and segmentation tasks. We obtained conclusive accuracy on the detection of GSS.