Visual versus region-of-interest based diffusion evaluation and their diagnostic impact in adult-type diffuse gliomas

To evaluate the comparability and reproducibility of standardized visual versus region-of-interest (ROI)-based diffusion assessment and their prediction capacity for isocitrate dehydrogenase (IDH) mutation status in adult gliomas.

Preoperative MRI scans, including diffusion-weighted imaging (DWI), of grade 2-4 adult-type diffuse gliomas (n = 303) were evaluated by three raters and repeated after one month. Visual assessment used the categorization of the Visually AcceSAble Rembrandt Images-feature 17 classes (facilitated, dubious, restricted). ROI-based assessment placed circular ROI on the visually perceived lowest apparent diffusion coefficient (ADC) areas (absolute/aADC) and contralateral normal-appearing white matter (normalized/nADC). Agreement and correlation analysis between visual and ROI-based assessments were performed. Logistic regression was conducted for IDH prediction in the subgroup of 99 non-necrotic and non-hemorrhagic cases, selected from the full cohort with available IDH status.

ROI-based assessment demonstrated superior inter- and intra-rater agreement (intraclass correlation coefficient[Formula: see text]0.56 (95%-CI: 0.48-0.63)) than visual assessment (Kendall's W/Cohen's weighted kappa[Formula: see text]0.34 (95%-CI: 0.26-0.42)). Thresholds of 1,090 and 623 × 10^-6 mm²/s for aADC, and 1.38 and 0.80 for nADC, distinguishing facilitated, dubious, and restricted diffusion, significantly correlated with visual assessments (P < .001). IDH classification accuracy of visual assessment was comparable to that of the ROI-based method using thresholds of aADC 1,048 × 10^- 6 mm²/sn and nADC 1.38 (visual vs. aADC/nADC: 69% vs. 73%/70%). However, neither method achieved a balanced performance between specificity (99% vs. 81%/75%) and sensitivity (14% vs. 57%/61%).

ROI-based diffusion assessment guided by visual input showed superior reproducibility than visual assessment alone. Although visual assessment demonstrated strong correlation with ADC thresholds and comparable overall IDH prediction accuracy, the two methods differ in clinical profile: visual assessment offered high specificity but low sensitivity, whereas ROI-based assessment improved sensitivity at the cost of reduced specificity.