Machine learning for triage of strokes with large vessel occlusion using photoplethysmography biomarkers

TL;DR

The paper addresses the challenge of rapidly identifying large vessel occlusion (LVO) stroke for direct transport to thrombectomy-capable centers. It evaluates a data-driven approach using 30-second fingertip photoplethysmography (PPG) biomarker features—morphological PPG features (MOR), beat-rate variability (BRV), and META demographics—to triage LVO vs non-LVO (NL) and stroke mimics (SM). The best-performing model, integrating MOR, BRV, and META features, achieves an AUROC of $0.77$ (0.71–0.82), outperforming single-modality models, with top features including a mix of META, MOR, and BRV signals. This study demonstrates the feasibility of rapid, noninvasive PPG-based triage in ambulance settings and motivates future work on multimodal data, ambulance-based data collection, and deeper learning approaches to improve prehospital stroke triage.

Abstract

Objective. Large vessel occlusion (LVO) stroke presents a major challenge in clinical practice due to the potential for poor outcomes with delayed treatment. Treatment for LVO involves highly specialized care, in particular endovascular thrombectomy, and is available only at certain hospitals. Therefore, prehospital identification of LVO by emergency ambulance services, can be critical for triaging LVO stroke patients directly to a hospital with access to endovascular therapy. Clinical scores exist to help distinguish LVO from less severe strokes, but they are based on a series of examinations that can take minutes and may be impractical for patients with dementia or those who cannot follow commands due to their stroke. There is a need for a fast and reliable method to aid in the early identification of LVO. In this study, our objective was to assess the feasibility of using 30-second photoplethysmography (PPG) recording to assist in recognizing LVO stroke. Method. A total of 88 patients, including 25 with LVO, 27 with stroke mimic (SM), and 36 non-LVO stroke patients (NL), were recorded at the Liverpool Hospital emergency department in Sydney, Australia. Demographics (age, sex), as well as morphological features and beating rate variability measures, were extracted from the PPG. A binary classification approach was employed to differentiate between LVO stroke and NL+SM (NL.SM). A 2:1 train-test split was stratified and repeated randomly across 100 iterations. Results. The best model achieved a median test set area under the receiver operating characteristic curve (AUROC) of 0.77 (0.71--0.82). \textit{Conclusion.} Our study demonstrates the potential of utilizing a 30-second PPG recording for identifying LVO stroke.

Figures (8)

• Figure 1: Graphical Abstract. First, the process begins with raw PPG extraction. The second step is preprocessing, which includes filtering and signal quality assessment. In the third step, features are extracted, including META features (age, sex), PPG morphological features (MOR), and beat rate variability (BRV). Fourth is the evaluation of feature importance, a machine learning-based model is trained and validated. Finally, in the fifth step, the triage of LVO stroke is tested and evaluated using area under the receiver-operating characteristics curve (AUROC). LVO: Large vessel occlusion stroke. NL.SM: non-LVO stroke + stroke mimics.
• Figure 2: Distribution of Hunter-8 scores. LVO: Large vessel occlusion stroke. NL.SM: non-LVO stroke + stroke mimics.
• Figure 3: Confusion Matrix for Hunter-8 Evaluation: This figure presents the classification performance of the Hunter-8 evaluation compared to confirmed diagnoses. The matrix highlights true positives, false positives, true negatives, and false negatives. Rows correspond to actual diagnoses, while columns represent predicted outcomes. LVO: Large vessel occlusion stroke. NL.SM: non-LVO stroke + stroke mimics.
• Figure 4: Sex (Panel A) and age (Panel B) distribution by stroke types. LVO: Large vessel occlusion stroke. NL: non-LVO stroke. SM: stroke mimics. NL.SM stroke includes both NL and SM cases.
• Figure 5: Classification of 88 patients. Patients’ recordings (N) were divided into 30-second windows (w). The train and test sets are stratified following signal quality checks. LVO: Large vessel occlusion stroke. NL: non-LVO stroke. SM: stroke mimics.