Comparison of Computerized and Vvisual Texture Analysis of Liver Focal Lesions in MRI

• Main Authors: A Gharbali, RA Lerski, SJ Gandy, R Bhat, P Clinch
• Format: Article
• Language: English
• Published: Tehran University of Medical Sciences 2005-10-01
• Series: Iranian Journal of Public Health
• Subjects: Texture analysis; Magnetic resonance imaging
• Online Access: http://journals.tums.ac.ir/PdfMed.aspx?pdf_med=/upload_files/pdf/2462.pdf&manuscript_id=2462

Even though each the focal liver lesions image has, it special pattern but in most of case differentiation between them is not easy task for radiologist. It seems computer aided differentiation can be useful in this step of diagnostic. Two independent radiologist assessed slice of MR liver images and twenty-three patients with focal liver lesions (3 Cyst, 6 Haemangioma and 14 Metastasis) and 10 normal livers were chosen for study. A texture analysis software and mathematical software were utilized to differentiate region of interest (ROI) among and in between ill and healthy liver slice images based on their differences in texture parameters. Linear discrimination analysis (LDA), Principle Component Analysis (PCA), combinations, and fusions of LDA and PCA were used as classification methods. Multiple ROIs were defined on control images to find out their best features data and linear discrimination functions for differentiation with high rate confidence. Sample images examined using the control set examination findings. The results then compared with radiologist reports. All classification methods allowed discrimination among and between healthy and focal lesion regions on the images. Automated texture analysis concurred with radiological diagnosis in all Cyst patients and all but one metastasis report. However, Haemangioma reports were classified as metastasis lesions. All samples of normal livers and normal parts of metastasis liver were correctly differentiated from metastasis. But more than 50% of patients reported as a metastasis diagnosed as normal. Comparison with visual diagnostic reports of MR liver images suggest that automated texture analysis has the potential to improve classification rates in the radiological diagnosis.