- Oral e-Poster Presentations - Booth 1: Vascular A (Aneurysms), September 25, 2023, 1:00 PM - 2:30 PM
Background: Subarachnoid hemorrhage(SAH) entails high morbidity and mortality. Several risk factors have been identified as mortality and functional outcome estimators. Artificial intelligence(AI) enables handling high-dimensional and complex data. Neural networks (NN), an automated machine learning technique, can be trained with images and/or data to perform accurate predictions. This study aims to predict the functional outcome of SAH patients at three months using a NN-based algorithm that processes initial CT scan images and clinical features.
Methods: Clinical features and CT scans of a multicentric retrospective cohort of SAH patients were analyzed. AUCMEDI, an open-source Python library, was used to create and train two different NNs: one based solely on images and the other incorporating clinical features (age and WFNS). The output variable was a dichotomized modifiedOral e-Poster Presentations - Booth 1: Vascular A (Aneurysms), September 25, 2023, 1:00 PM - 2:30 PM
Background: Subarachnoid hemorrhage(SAH) entails high morbidity and mortality. Several risk factors have been identified as mortality and functional outcome estimators. Artificial intelligence(AI) enables handling high-dimensional and complex data. Neural networks (NN), an automated machine learning technique, can be trained with images and/or data to perform accurate predictions. This study aims to predict the functional outcome of SAH patients at three months using a NN-based algorithm that processes initial CT scan images and clinical features.
Methods: Clinical features and CT scans of a multicentric retrospective cohort of SAH patients were analyzed. AUCMEDI, an open-source Python library, was used to create and train two different NNs: one based solely on images and the other incorporating clinical features (age and WFNS). The output variable was a dichotomized modified Rankin scale at 3 months(mRS): Good Outcome=mRS<4; Bad Outcome= mRS<4. The initial dataset was randomly split into training, validation, and test cohorts at a ratio of 70%-10%-20%.
Results: Images and data from 219 patients were processed. 52.5% were female patients with a mean age of 57. 18.3% were idiopathic SAH. Median WFNS on admission was 2, and mortality was 28.8%. 54.3% of patients presented a good outcome at 3 months follow-up. Predicting neurological outcome, the model exclusively based on CT scan images (Accuracy=86%, F1=86% and AUC=0.89) outperformed the one based on images and clinical data (Accuracy=79%, F1=78% and AUC=0.87). Explainable Artificial Intelligence maps were built to highlight the areas the algorithm accounts on the CT scan in order to classify patients.
Conclusions: Modern image processing techniques based on AI and NN make possible to predict neurological outcome in SAH patients with high accuracy using CT scan images as the only input. Models might be optimized by including more data and patients, therefore improving their performance on tasks beyond the skills of conventional clinical knowledge.…
