Diffuse Optical Tomography (DOT) provides quantitative information about optical absorption and total hemoglobin concentration of breast tumors which is directly related to tumor angiogenesis. However, measurements errors caused by tissue heterogeneity may introduce artifacts in reconstructed absorption and total hemoglobin maps and therefore cause errors in quantitative characterization of lesions. With ultrasound-guided DOT, these artifacts can be recognized if they are isolated and located at edges of the absorption maps. However, effectively recognize image artifacts and automatically remove them is a challenge because artifacts can be merged partially with lesions maps. A new two-step algorithm is proposed to iteratively identify and remove measurement outliers based on assignment of local outlier factors and hence to reduce image artifacts and produce more consistent absorption maps among different wavelengths. In first step, perturbation measurements were ranked based on data density and the local outlier factor which is the probability of each measurement being an outlier. In second step, outliers are iteratively removed until normalized pattern correlation of different wavelength absorption maps is beyond a specified threshold. The proposed algorithm is evaluated on 20 clinical cases and it has demonstrated its capability to automatically reconstruct more consistent images for different wavelengths. The improvement on characterizing benign breast lesions is more dramatic because outliers can cause the reconstructed benign lesions with higher optical absorption and therefore high hemoglobin contrast.
|