Fig. 4
From: High-throughput quantitative histology in systemic sclerosis skin disease using computer vision
Prediction of fibrosis and association with molecular pathology. A) We developed and independently tested a linear regression model composed of Quantitative Image Features (QIFs) to predict mRSS. The model output was a Fibrosis Score measured arbitrary units (AU). Using cross validation (CV) on the training data set, the Fibrosis Score strongly correlated with modified Rodnan Skin Score (mRSS) (R = 0.70, open squares). This correlation generalized to the independent test set (R = 0.55, p = 5.3 × 10− 6; solid dots). B) The DNN-derived Fibrosis Score significantly correlated with the validated Scleroderma Skin Severity Score (4S) (R = 0.69, p = 2.9 × 10− 17)