Treatment Efficacy Analysis in Acute Ischemic Stroke Patients Using In Silico Modeling Based on Machine Learning: A Proof-of-Principle

Abstract

Interventional neuroradiology is characterized by engineering- and experience-driven device development with design improvements every few months. However, clinical validation of these new devices requires lengthy and expensive randomized controlled trials. This contribution proposes a machine learning-based in silico study design to evaluate new devices more quickly with a small sample size. Acute diffusion- and perfusion-weighted MRI, segmented one-week follow-up imaging, and clinical variables were available for 90 acute ischemic stroke patients. Three treatment option-specific random forest models were trained to predict the one-week follow-up lesion segmentation for (1) patients successfully recanalized using intra-arterial mechanical thrombectomy, (2) patients successfully recanalized using intravenous thrombolysis, and (3) non-recanalizing patients as an analogue for conservative treatment for each patient in the sample, independent of the true group membership. A repeated-measures analysis of the three predicted follow-up lesions for each patient revealed significantly larger lesions for the non-recanalizing group compared to the successful intravenous thrombolysis treatment group, which in turn showed significantly larger lesions compared to the successful mechanical thrombectomy treatment group (p < 0.001). A groupwise comparison of the true follow-up lesions for the three treatment options showed the same trend but did not reach statistical significance (p = 0.19). We conclude that the proposed machine learning-based in silico trial design leads to clinically feasible results and can support new efficacy studies by providing additional power and potential early intermediate results.

Posterior Circulation Endovascular Thrombectomy for Large Vessels Occlusion in Patients Presenting with NIHSS Score ≤ 10

“…Our study presents several limitations: first, the design of a retrospective observational study. Hypothesize stroke's etiology, as well as other parameters (e.g., mTICI and ASPECT score) were assessed by both the neurologist and interventional neuroradiologist without a central core-lab, or without using an automated imaging software analysis [38,39]. Consequently, the possibility of selection bias cannot be excluded.…”

Predicting the tissue outcome of acute ischemic stroke from acute 4D computed tomography perfusion imaging using temporal features and deep learning

“…The different tissue outcome prediction models were separately applied and evaluated using only the patients treated with intra-arterial mechanical thrombectomy (IA), and then using only the patients treated with intravenous thrombolysis (IV). This approach guarantees functionally independent predictive models for different treatment approaches, which is important for many potential applications of tissue outcome prediction such as treatment efficacy analysis (Fiehler et al, 2019;Winder et al, 2021). The deep learning models are also visualized in Figure 1.…”

“…Applied to a population of patients, such an ensemble of machine learning models trained on a rather small set of datasets could help to determine if a novel treatment devices or therapy has the potential for a full-scale clinical trial based on its predicted tissue salvage. This is referred to as in silico treatment efficacy analysis (Winder et al, 2021). Applied to individual patients, a similar ensemble could help clinicians to understand how a particular case of stroke will evolve without medical intervention and then balance the potential tissue salvage of different treatment options with their respective financial costs and health risks.…”

“…Significant differences between the IA and IV cohorts do not introduce bias into a comparison of different model architectures, considering that each architecture was trained and evaluated on the same patient data. However, any comparison between one model trained on the IA cohort and another model trained on the IV cohort would be subject to this bias, which limits the applicability of the models in their current state to tasks such as in silico treatment efficacy analysis (Winder et al, 2021). Furthermore, our models are agnostic to potentially relevant clinical variables such as those listed at the end of Section "Effects of deconvolution and auxiliary inputs" and, as mentioned in Section "Current and future applications of machine learning-based stroke tissue outcome prediction, " further experimentation is required to integrate these variables.…”

“…In the context of validating novel therapeutic devices, it has been proposed that tissue outcome predictions could be 10.3389/fnins.2022.1009654 used as a virtual endpoint for in silico clinical trials (Winder et al, 2021). The ERASER clinical trial, for example, validated a novel stent retrieval device using random forests to show a reduced predicted infarct volume for IA treatment compared to the true IV treatment outcome (Fiehler et al, 2019).…”