Appearance analysis for diagnosing malignant lung nodules

Abstract

An alternative method of diagnosing malignant lung nodules by their visual appearance rather than conventional growth rate is proposed. Spatial distribution of image intensities (or Hounsfield values) comprising the malignant nodule appearance is accurately modeled with a rotation invariant second-order Markov-Gibbs random field. Its neighborhood system and potentials are analytically learned from a training set of nodule images with normalized intensity ranges. Preliminary experiments on 109 lung nodules (51 malignant and 58 benign ones) resulted in the 96.3% correct classification (for the 95% confidence interval), showing the proposed method is a promising supplement to current technologies for early diagnostics of lung cancer.