In a study of 66 patients who suffered an acute ischemic stroke, researchers found that adding brain images taken on the first day after admission improved the statistical ability of computer models to predict patients’ functional status three months later. The models used deep‑learning image representations from four MRI sequences (ADC, DWI, FLAIR, and the corresponding clinical data). When the models combined the different sources of information, the area under the receiver operating characteristic curve (AUC) rose to 0.948 ± 0.061, a 0.049‑point boost over a model that used only clinical variables. The strongest single‑modal image contribution came from FLAIR, which, when fused with the clinical variables and ADC‑derived lesion biomarkers, produced the highest discrimination. Models that used all three imaging modalities together did not perform better than the best single‑modality combination. Despite or because of this, the model that used only structured clinical data (including age, NIHSS scores, and other standard measures) achieved the highest accuracy (0.845 ± 0.161) and F1‑score (0.820 ± 0.145). The study employed class‑balanced linear support‑vector machines and evaluated performance with eight‑fold stratified group cross‑validation at the patient level. The research highlights that while image‑based features provide statistically significant improvement, the traditional clinical assessment remains critical for accurate outcome prediction. However, the authors caution that the sample size is small and the data come from a single center, so external validation is needed before these models can be applied broadly.