A multiparameter magnetic resonance imaging–based radiomics model could predict progressive neurologic deterioration in patients with subacute ischemic perforator artery cerebral infarction with improved predictive performance compared with single-sequence models, achieving an area under the curve of 0.844 in the training cohort and 0.824 in the validation cohort.
In a retrospective two-center study, investigators evaluated whether radiomic features extracted from T2-weighted imaging, single-shot echo planar imaging diffusion-weighted imaging, and apparent diffusion coefficient maps could predict progressive cerebral infarction, defined as an increase of 2 or more points on the National Institutes of Health Stroke Scale during hospitalization compared with admission.
The investigators included 188 patients with single-lesion subacute ischemic perforator artery cerebral infarction who underwent magnetic resonance imaging (MRI) within 7 days of symptom onset. The patients were randomly assigned in a 7:3 ratio to a training cohort (n = 131) and a validation cohort (n = 57). The investigators noted that 66 of the patients met the criteria for progressive cerebral infarction, whereas 122 of them were classified as nonprogressive.
Radiologists manually delineated the largest lesion layer on each sequence, and 1,320 radiomic features were extracted per sequence following voxel resampling to 1 × 1 × 1 mm³ and preprocessing. Feature selection incorporated independent sample T-testing, recursive feature elimination, and least absolute shrinkage and selection operator regression with fivefold cross-validation. Six optimal features were retained for T2-weighted imaging, four for diffusion-weighted imaging, and 11 for apparent diffusion coefficient maps.
Multivariable logistic regression demonstrated that radiomic scores derived from diffusion-weighted imaging, apparent diffusion coefficient, and T2-weighted imaging were independently associated with progression; whereas admission National Institutes of Health Stroke Scale scores, C-reactive protein levels, and systolic blood pressure weren't statistically significant in the combined analysis.
Using a support vector machine classifier, the combined model integrating all three sequences outperformed individual sequence models. In the training cohort, the combined model achieved about 76% sensitivity, 75% specificity, and 76% accuracy. In the validation cohort, sensitivity was about 65%, specificity 73%, and accuracy 70%. By comparison, single-sequence models yielded area under the curve values ranging from 0.769 to 0.820 in the training cohort and 0.768 to 0.791 in the validation cohort. Decision curve analysis demonstrated greater net benefit across threshold probabilities for the combined model, and calibration curves showed concordance between predicted and observed outcomes.
The investigators noted several limitations, including the modest sample size, restriction to single perforating artery infarctions, heterogeneity in infarct location, lack of external validation, and retrospective design. They also acknowledged potential partial volume effects and limited follow-up regarding postinfarction edema and hemorrhagic transformation.
“In summary, we developed a radiomics-based model using [T2-weighted imaging], [diffusion-weighted imaging], and [apparent diffusion coefficient] images to predict progression of clinical symptoms in patients with subacute ischemic [perforator] artery cerebral infarction,” noted lead study author Wenjing Yu, of the Department of Radiology at The First Affiliated Hospital of Zhejiang Chinese Medical University (Zhejiang Provincial Hospital of Chinese Medicine) as well as The First School of Clinical Medicine of Zhejiang Chinese Medical University in China, and colleagues.
The researchers reported no conflicts of interest.
Source: BMC Medical Imaging
