1379 Machine Learning Models Incorporating Clot Histomics and Transcriptomics Predict Functional Outcome After Mechanical Thrombectomy in Acute Ischemic Stroke

INTRODUCTION: Histologic and transcriptomic analyses of retrieved stroke clots have identified features associated with patient outcomes, but few studies have investigated how paired histologic image features (histomics) and expression patterns from the same clot can improve biological interpretation and outcome prediction. METHODS: We performed paired histological and transcriptomic analysis of 32 stroke clots. ENI was defined as a delta-NIH Stroke Scale>4, and a good long-term outcome was defined as mRS=2 at 90 days post thrombectomy. 40X H&E clot whole-slide images were analyzed with an established digital pathology pipeline, extracting 237 histomics and %Comp. Differentially expressed genes (DEGs) were identified as those with absolute fold-change>1.5 and q<0.05 between ENI or mRS outcomes. Machine learning with recursive feature elimination and cross-validation was used to select clot features and develop prediction models. RESULTS: 9 ENI DEGs and 3 mRS DEGs were identified. While ENI DEGs highlighted neutrophil-mediated inflammation, mRS DEGs highlighted purine nucleobase salvage towards restoration of cerebral ATP. Feature selection identified a combination of 9 histologic and transcriptomic features that predicted ENI with an accuracy of 90.8% and AUC=0.98±0.08 and 7 histomic and transcriptomic features that predicted 90-day mRS with an accuracy of 93.7% and AUC=0.94±0.09. Overall, the hybrid histomic transcriptomic models predicted functional outcomes better than models based on histomics, transcriptomics, or %Comp alone. CONCLUSIONS: Hybrid models offer improved outcome prediction for patients with stroke. Identified digital histology and mRNA signatures warrant further investigation as biomarkers of patient functional outcome after thrombectomy.