Discriminating Malignant and Benign Brain Tumors Using Texture Features Of MRI-ADC Images

Abstract

Introduction: The diagnosis of brain tumors often involves the use of Magnetic Resonance Imaging (MRI), with the Apparent Diffusion Coefficient (ADC) being a commonly employed technique in current clinical practice. This study seeks to investigate the potential of using statistical texture analysis of MRI-ADC images to distinguish between malignant and benign brain tumors. Methods: The research utilized 980 MRI brain ADC image slices labeled as malignant and 805 labeled as benign from 252 subjects. The clinical diagnosis of each participant was verified by histopathological and radiological reports. The region of interest (ROI) was defined to extract ADC values within the tumor areas. From each ROI, statistical features including higher-order moments of ADC, mean pixel value, and texture features of Grey Level Co-occurrence Matrix (GLCM) were extracted along with patient demographic information. The mean feature values for each category were computed and analyzed using a one-tailed P value test at a 95% confidence level. Results: The average pixel value of ADC, as well as the GLCM texture features (Variance 1, Variance 2, Mean 1, Mean 2, Contrast, and Energy), were found to be significantly higher (P<0.05) for benign tumors. In Contrast, malignant tumors exhibited significantly higher values for kurtosis of ADC and GLCM texture features (Entropy, Homogeneity, and Correlation). The patient's age and other features (skewness of ADC, GLCM texture features such as Shade, Entropy, and Prominence) did not provide sufficient evidence to reject the null hypothesis (P>0.05). Conclusions: In conclusion, the aforementioned features, with the exception of the patient's age, skewness, and GLCM features such as Entropy, Shade, and Prominence can be used as potential biomarkers for distinguishing between benign and malignant brain tumors.