Skip to main content

Table 3 Studies investigating the use of MRI texture analysis in diagnosis, treatment response assessment, and as a prognostic tool

From: Assessment of tumor heterogeneity: an emerging imaging tool for clinical practice?

Diagnosis and characterization Method Study findings Author, year
Simulated microcalcification Texture analysis Successful automatic detection of localized blurring was achieved (sensitivity = 89 %–94 %; specificity = 99.7 %―100 %; PPV = 74 %–100 %; NPV = 99.9 %–99.9 %) James et al., 2001 [45]
Breast cancer Texture analysis A combination of textural analysis (second-order statistics, e.g., contrast, sum entropy, entropy), lesion size, time to maximum enhancement, and patient age allowed for a diagnostic accuracy of 0.92 ± 0.05 Gibbs et al., 2003 [46]
Breast lesion Texture analysis The classification performance of volumetric texture features (second-order statistics) is significantly better than 2D analysis Chen et al., 2007 [47]
Breast cancer Texture analysis The 4D texture analysis (using second-order statistics) achieved a performance comparable to human observers Woods et al.,2007 [48]
Invasive lobular and ductal breast cancer Texture analysis Investigated the use of first-order statistics, second-order stastistics obtained from GLCM, RLM, autoregressive model, and wavelet transform. All parameters distinguished healthy from cancerous tissue although GLCM performed better. 80 %–100 % of accuracy in differentiating ductal from lobular cancers, particularly complexity and entropy Holli et al., 2010 [22]
Glioneuronal tumor Texture analysis The combination of DCE-MRI and MRI textural analysis (second-order statistics—GLCM and RLM) provide optimal differentiation between glioneuronal tumors and gliomas in vivo Eliat et al., 2012 [19]
Brain tumors—metastases, meningiomas, gliomas (grade II and III), glioblastomas Texture analysis Metastases were successfully distinguished from gliomas (accuracy = 85 %; sensitivity = 87 %; specificity = 79 %) as well as high-grade from low-grade neoplasms (accuracy = 88 %; sensitivity = 85 %; specificity = 96 %) using Gabor transform texture analysis Zacharaki et al., 2009 [23]
Prostate cancer Fractal analysis The combination of fractal and multifractal features was more accurate than classical texture features in detecting cancer and was more robust against signal intensity variations Lopes et al., 2011 [21]
Prostate cancer Fractal analysis Both fractal analyses offered promising quantitative indices for prostate cancer identification, with histogram fractal dimension offering a more robust diagnosis than texture fractal analysis (correlation coefficient of c = 0.9905 vs. c = 0.9458, respectively) Lv et al., 2009 [49]
Liver cysts and hemangiomas Texture analysis Texture analysis (first-order, second-order statistics and wavelet transform) was successfully used to classify focal liver lesions on zero-fill interpolated 3.0-T MR images Mayerhoefer et al., 2010 [20]
Response assessment
Breast Texture analysis Second-order statistics extracted from parametric maps that reflect lesion washout properties discriminate malignant from benign tumors better than textural features extracted from either first post-contrast frame lesion area or from parametric map reflecting lesion initial uptake. Angular second moment and entropy were most discriminative Karahaliou et al., 2010 [26]
Lymphoma Texture analysis Texture analysis [first-order, second-order statistics (GLCM and RLM), autoregressive model and wavelet transform] was able to classify NHL lesions undergoing chemotherapy based on changes following treatment Harrison et al.,2009 [27]
Liver metastases Fractal analysis Tumor heterogeneity as assessed by fractal dimension predicted tumor shrinkage in response to bevacizumab and cytotoxic chemotherapy in colorectal liver metastases O’Connor et al., 2011 [28]