Automatic Detection of Image-Based Features for Immunosuppressive Therapy Response Prediction in Oral Lichen Planus

Oral lichen planus (OLP) is a chronic inflammatory disease, and the common management focuses on controlling inflammation with immunosuppressive therapy. While the response to the immunosuppressive therapy is heterogeneous, exploring the mechanism and prediction of the response gain greater importance. Here, we developed a workflow for prediction of immunosuppressive therapy response prediction in OLP, which could automatically acquire image-based features. First, 38 features were acquired from 208 OLP pathological images, and 6 features were subsequently obtained which had a significant impact on the effect of OLP immunosuppressive therapy. By observing microscopic structure and integrated with the corresponding transcriptome, the biological implications of the 6 features were uncovered. Though the pathway enrichment analysis, three image-based features which advantageous to therapy indicated the different lymphocytes infiltration, and the other three image-based features which bad for therapy respectively indicated the nicotinamide adenine dinucleotide (NADH) metabolic pathway, response to potassium ion pathway and adenosine monophosphate (AMP) activated protein kinase pathway. In addition, prediction models for the response to immunosuppressive therapy, were constructed with above image-based features. The best performance prediction model built by logistic regression showed an accuracy of 90% and the area under the receiver operating characteristic curve (AUROC) reached 0.947. This study provided a novel approach to automatically obtain biological meaningful image-based features from unannotated pathological images, which could indicate the immunosuppressive therapy in OLP. Besides, the novel and accurate prediction model may be useful for the OLP clinical management.