Expectation-maximization-estimation of mixture densities for Electron-Spin-Resonance-analysis of albumin

摘要

Early diagnosis of human cancer is of crucial importance for successful therapies. Cancer diagnosis via ESR (electron-spin-resonance) spectroscopy of albumin found in human blood provides a new promising approach. The ESR frontend signal processing follows a protocol of our proprietary dasiamobility of molecular structure testpsila (MMS-Test) and provides a real-valued 33-dimensional vector representation per sample, which combines a representative feature set of the binding ability (spin-probes) of albumin under investigation. Classical statistical pattern recognition is then applied to the feature vector, including LDA and EM mixture density estimation, leading to a classification error rate of 14% between the two patient classes dasiahealthypsila and dasiasuspectpsila. The class dasiasuspectpsila includes cancer and other chronic condition. The investigation was performed on a proprietary database of MedInnovation with 1176 cancer and non-cancer patients.