top of page

Diffusion tensor imaging radiomics in lower-grade glioma: improving subtyping of isocitrate dehydrog

Figure 1. Workflow to image processing to machine learning. First, postcontrast T1 (T1C), T2 and FLAIR images and ADC map are registered to the same spatial coordinates. After signal intensity normalization of T1C, T2, and FLAIR images, radiomic features are extracted, including 253 features from conventional images, and 158 features from ADC map. These features are subjected to the Random Forest (RF) modelling to predict IDH mutation status. To mitigate overfitting from small sample size, nested cross validation was conducted.

Glioma is the most common primary brain tumor in adults. A mutation in isocitrate dehydrogenase (IDH) is a major classifier of diffuse gliomas, with IDH wild-type lower-grade gliomas (LGGs) having worse prognosis compared with IDH mutated lower-grade gliomas. Preoperative imaging biomarkers have been investigated for noninvasive subtyping of IDH status, including radiomic analysis. Recently introduced radiomics enables the extraction of high-dimensional quantitative features that incorporates spatial interrelationships of signal intensity from medical images. Apparent diffusion coefficient (ADC) and fractional anisotropy (FA) from diffusion tensor imaging (DTI) are known as an index of tumor cellularity and white matter integrity and has been reported to be correlated with grade and molecular subtype, such as IDH status of gliomas. However, the role of radiomic features from DTI is not established for IDH status prediction in LGGs. The purpose of our study was to evaluate whether machine learning–based radiomics analysis of DTI improves the subtyping of IDH mutation status in lower-grade gliomas beyond than that of radiomic features from conventional MRI and DTI histogram parameters.

A total of 168 patients with pathologically confirmed lower-grade gliomas were retrospectively enrolled. A total of 158 and 253 radiomic features were extracted from DTI (DTI radiomics) and conventional MRI (T1-weighted image with contrast enhancement, T2-weighted image, and FLAIR [conventional radiomics]), respectively (Figure 1). The random forest models for predicting IDH status were trained with variable combinations as follows: (1) DTI radiomics, (2) conventional radiomics, (3) conventional radiomics + DTI radiomics, and (4) conventional radiomics + DTI histogram. The models were validated with nested cross validation. The predictive performances of those models were compared by using area under the curve (AUC) from receiver.

Results are shown in Figure 2. Adding DTI radiomics to conventional radiomics significantly improved the accuracy of IDH status subtyping (AUC, 0.900, p = 0.006), whereas adding DTI histogram parameters yielded nonsignificant trend toward improvement (0.869, p = 0.150) compared with the model with conventional radiomics alone (0.835). The performance of the model consisting of both DTI and conventional radiomics was significantly superior than that of model consisting of both DTI histogram parameters and conventional radiomics (0.900 vs 0.869, p = 0.040). Total 18 features were selected among 158 DTI radiomic features to be relevant for IDH status prediction. (Figure 3)

Clinical implication:

Although radiomics provides comprehensive quantitative information, radiomics analysis requires more complicated preprocessing and computation than conventional imaging analysis. Thus, it is of interest to determine whether this complicated process of radiomics is worthy. Our results suggest that those labor-intensive radiomics analysis using DTI is worthy, and enables further exploitation of given DTI and improves the prediction of IDH status. Additionally, as DTI does not require contrast enhancement, DTI radiomics has the potential to provde further information regardless of contrast administration feasibility.

조회수 66회댓글 0개
bottom of page