Improved automated localization and quantification of protein multiplexes via multispectral fluorescence imaging in heterogenous biopsy samples

We present a novel improvement of our previously published image analysis system for the automated localization and quantification of protein biomarker expression in immunofluorescence (IF) microscopic images. The improvement has been developed primarily for biopsy based images which are by nature of variable quality and heterogeneous. The innovative method is employed for discriminating the biomarker signal from background, where signal may be the expression of multiple biomarkers or counterstains used in IF. The method is dynamic and it derives a threshold for a true biomarker signal based on the relationship between disease and non-disease tissue components. In addition, a new dynamic range feature construction is presented that is less affected by processing and other variations in tissue. The utility of the approach is demonstrated in predicting, based on the diagnostic biopsy tissue, prostate cancer disease progression within eight years after a radical prostatectomy. For this purpose, androgen receptor (AR) and Ki67 biomarker expression in prostate biopsy samples was quantified and features from the proposed approach were shown to be associated with disease progression in a univariate analysis and manifested improved performance over prior systems. Furthermore, AR and Ki67 features were selected in a multivariate model integrating clinical, histological, and biomarker features, proving their independent prognostic value.