Integrating DWI-derived intra-infarction radiomics, 2 mm peri-infarction signatures, and clinical indicators achieves optimal AUC discrimination within the training cohort, though its statistical advantage over isolated intra-infarction radiomics could not be validated in the independent test set.
Glucose level was an independent clinical predictor of neurological deterioration in acute ischemic stroke.
The XGBoost-based intra-infarction radiomics model achieved the best predictive performance among intra-infarction models.
The 2 mm peri-infarction radiomics model performed best among all peri-infarction models.
The combined model integrating intra-infarction radiomics, 2 mm peri-infarction radiomics, and clinical features achieved the highest AUC values overall.
The combined model achieved a perfect sensitivity of 1.000 in the test cohort, but this finding requires cautious interpretation due to the small number of ND events.
Four predictive model types were constructed and compared: clinical, intra-infarction radiomics, peri-infarction radiomics, and a combined model.
This research suggests that analyzing MRI brain scan features from both the core area of a stroke and a small surrounding zone (called the peri-infarction region) can help predict which stroke patients will experience worsening of their neurological condition — a complication called neurological deterioration. The researchers used a type of MRI called diffusion-weighted imaging (DWI) and extracted hundreds of mathematical image features (radiomics) from the stroke area and a 2 mm ring of tissue around it in 767 patients. They then trained computer models to predict which patients would deteriorate, finding that the best single-region approach used the stroke core features with an XGBoost algorithm (achieving roughly 90% accuracy on new patients), while combining core features, surrounding tissue features, and blood glucose levels gave the highest overall accuracy (about 91% on new patients). However, the study also highlights important limitations. Although the combined model detected every case of neurological deterioration in the test group (100% sensitivity), there were only 25 deterioration events in that group, making this result statistically fragile. Additionally, the improvement of the combined model over using just the stroke core features alone was not statistically significant when tested on new patients, meaning the added value of including the surrounding tissue and clinical data could not be definitively confirmed. Blood glucose was the only clinical measurement independently linked to worsening outcomes. This research suggests that advanced image analysis of stroke MRI scans — particularly from both the damaged tissue and its immediate surroundings — could one day help doctors identify high-risk patients earlier and more objectively than current methods allow. However, the authors emphasize that larger studies involving multiple hospitals are needed before this approach can be broadly applied in clinical practice, as the current findings may not generalize to all stroke patients.
Si P, Zong C, Liu L, Wang D, Song B. (2026). Development and validation of a DWI-based intra- and peri-infarction radiomics model for predicting neurological deterioration in acute ischemic stroke.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1825520