Work

Nimg-13quantitative Mr Texture Analysis Can Identify Low Grade Glioma and Peritumoral White Matter From Normal Appearing White Matter (nawm) Phenotypes

Year: 2015

Type: article

Abstract: BACKGROUND: To assess the efficacy of texture analysis (TA) in differentiating Low Grade Gliomas (LGGs) from Normal Appearing White Matter (NAWM) in a cohort of brain cancer patients using T2-FLAIR MR images. Also the impact of the number of gray-levels employed was investigated. METHODS: T2-FLAIR and post-gadolinium (GD) contrast T1-weighted MR images of 11 TCGA patients (mean age 40.8 years, range 27-59.5 years) were retrospectively analyzed. All patients were diagnosed with LGGs (grades II and III). Four different volumes of interest (VOI)s were segmented for each patient by an experienced radiologist: i) tumor, ii)peritumoral WM (pWM), iii) far ipsi-lateral WM (iWM) and iv) contra-lateral WM(cWM). Following VOI segmentation, TA was performed using FLAIR images on MatLab environment. A complete list of invariant texture features derived from the gray-level co-occurrence matrix (GLCM) was calculated. Because the ideal number of gray levels is unknown, we used various gray levels (8, 16, 32, and 64 levels). To assess the efficacy of the calculated features we constructed a decision tree. The model was validated using a 10-fold cross validation of the training data. The accuracy, sensitivity and specificity were calculated. Impact of the number of gray levels on the performance indicators was investigated. RESULTS: For discrimination of LGG from NAWM, TA achieved 100% accuracy. High accuracy (80-87.5% for different gray levels) was achieved for discrimination of all four VOIs. Off note, the peritumoral WM was identified with 86.7% to 93% accuracy from the two remaining NAWM regions combined. Also, from the performance indicators it is evident that the number of gray levels does not affect the calcification. CONCLUSIONS: TA provided excellent accuracy for discrimination between LGG and NAWM. A major strength of the presented work is that it showed high performance in discriminating among normal WM regions depending on their proximity to the tumor.

Source: Neuro-Oncology

DOI: 10.1093/neuonc/nov225.13

Authors: Aikaterini Kotrotsou, Pascal O. Zinn, Masumeh Hatami, Rivka Colen

Institutions: The University of Texas MD Anderson Cancer Center, Baylor College of Medicine

Language: English

Topic: Radiomics and Machine Learning in Medical Imaging

Subfield: Radiology, Nuclear Medicine and Imaging

Field: Medicine

Domain: Health Sciences

SDG: Good health and well-being

Open Access status: bronze

OpenAlex ID: W2584951907