Validating the Benefit of Combining Imaging and Clinical Data for Ischemic Stroke Outcome Prediction

Zeyad Abouyoussef, Roberto Souza, Johanna Ospel

Keywords: ischemic stroke, deep learning, outcome prediction, multimodal learning.

Abstract:

Endovascular treatment has been proven to be a successful treatment for acute ischemic stroke. However, the selection criteria for this procedure is oversimplified and based on a small number of factors. This leads to the exclusion of patients who would have benefited from treatment and the inclusion of patients who would not have benefited from treatment. Additionally, radiologists rely on their expertise in identifying infarcted tissue, and with the amount of information to consider, it can be overwhelming to process all of the information at decision making time. An automated way of selecting patients for treatment and predicting how they respond to treatment in the long term can aid radiologists and stroke physicians in their decision-making.

Deep learning has been used in predicting the functional outcome from pre-treatment imaging and clinical data, with most studies combining either non-contrast or angiography computed tomography (CT) scans with clinical data. However, there is no conclusive evidence that deep learning is superior to traditional machine learning models trained using tabular clinical data. Training with only one imaging modality can be a contributing factor to the models not learning a complete feature representation of the stroke case. Exploring the effect of training using all the available information can be helpful in boosting deep learning model performance.

We investigate the effect of training multimodal deep learning models using imaging and clinical data on the same set of patients. This will lead to a better understanding of the added benefit of each imaging modality. Clinically, better model performance leads to better estimates and a more refined selection criterion for treatment.

Primary Subject Area: Integration of Imaging and Clinical Data

Secondary Subject Area: Application: Neuroimaging

Registration Requirement: Yes

Reproducibility: https://github.com/zeyad-kay/multimodal_mrs_prediction

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 340