Differentiation of brain metastases in MRI image using the first and second-order statistical features

Abstract

The brain is highly susceptible to metastases from lung cancer. The segmentation and detection of brain metastases are the main goals for the management of patients with brain metastases and a MRI technique that uses the image signal contrast between tissues rather than their absolute signal intensities is a recommended approach. This paper proposes a specific quantification method for the most relevant first and second-order features computed for brain images acquired as T2‐weighted and PD (proton density). T2-w sequences are useful for detecting high-signal tumor infiltration whilst PD (proton density) sequences performed a fat suppression being an intermediate sequence between T1-w and T2-w and share common features of both. Based on the first-order histogram (the gray-level distribution of the image) and on texture-related information provided by the co-occurrence matrix, features like skewness, kurtosis or entropy, and energy were computed in terms of their discriminatory power for a better clinical investigation of MRI images. Our analysis provides for a smaller number of relevant and distinguishable features and the computational task is at a reasonable level.