Towards MIB-1 and p53 detection in glioma magnetic resonance image: a novel computational image analysis method.

Glioma is the primary tumor in the central nervous system, and poses one of the greatest challenges in clinical treatment. MIB-1 and p53 are the most useful biomarkers for gliomas and could help neurosurgeons establish a therapeutic schedule. However, these biomarkers are commonly detected with the help of immunohistochemistry (IHC), which wastes time and energy and is often influenced by subjective factors. To reduce the subjective factors and improve the efficiency in the judgment of IHC, a novel magnetic resonance image (MRI) analysis method is proposed in the present study to detect the expression status of MIB-1 and p53 in IHC. The proposed method includes two kinds of MRI acquisition (FLAIR and T1 FLAIR images), regions of interest (ROIs) selection, texture features (i.e. the gray level gradient co-occurrence matrix (GLGCM), Minkowski functions (MFs), etc) extraction in ROIs, and classification with a support vector machine in a leave-one-out cross validation strategy. By classifying the ROIs, the performance of the method was evaluated by accuracy, area under ROC curve (AUC), etc. A high accuracy (0.7640 ± 0.0225) and AUC (0.7873 ± 0.0377) for MIB-I detection were achieved. In terms of the texture features, 0.7621 ± 0.0199, 0.7666 ± 0.0365 and 0.7426 ± 0.0451 AUC can be obtained using only GLCM, RLM or GLGCM for MIB-1 detection, respectively. In all, the experimental results demonstrated that MR image texture features are associated with the expression status of MIB-1 and p53. The proposed method has the potential to realize high accuracy and robust detection for MIB-I expression status, which makes it promising for clinical glioma diagnosis and prognosis.