Comparing Baseline and Day-1 Diffusion MRI Using Multimodal Deep Embeddings for Stroke Outcome Prediction

TL;DR

This study directly compares baseline (J0) and day-1 (J1) diffusion MRI for predicting three-month outcomes after acute ischemic stroke, using a multimodal deep-embedding framework that fuses ADC-derived 3D-ResNet features with structured clinical data and lesion-volume metrics. J1 diffusion provides a stronger prognostic signal, with MRI-only AUC of $0.714 \pm 0.105$ (vs $0.540 \pm 0.263$ for J0) and a multimodal J1 model achieving AUC $0.923 \pm 0.085$ when including lesion-volume features. Dimensionality reduction via PCA (≤12 components) and a linear SVM yield a transparent, data-efficient classifier that benefits from lesion-volume integration for stability and interpretability. The results support using early post-treatment diffusion MRI to improve personalized AIS prognosis and present a reproducible PCA–SVM fusion approach suitable for clinical translation, with validation in larger multicenter cohorts planned.

Abstract

This study compares baseline (J0) and 24-hour (J1) diffusion magnetic resonance imaging (MRI) for predicting three-month functional outcomes after acute ischemic stroke (AIS). Seventy-four AIS patients with paired apparent diffusion coefficient (ADC) scans and clinical data were analyzed. Three-dimensional ResNet-50 embeddings were fused with structured clinical variables, reduced via principal component analysis (<=12 components), and classified using linear support vector machines with eight-fold stratified group cross-validation. J1 multimodal models achieved the highest predictive performance (AUC = 0.923 +/- 0.085), outperforming J0-based configurations (AUC <= 0.86). Incorporating lesion-volume features further improved model stability and interpretability. These findings demonstrate that early post-treatment diffusion MRI provides superior prognostic value to pre-treatment imaging and that combining MRI, clinical, and lesion-volume features produces a robust and interpretable framework for predicting three-month functional outcomes in AIS patients.

Figures (5)

• Figure 1: Thresholding of lesion-like regions in ADC maps at different intensity levels. (Left) Raw slice, (Middle) Threshold $<$ 620, (Right) Threshold $<$ 480.
• Figure 2: Overview of the proposed multimodal pipeline combining clinical data and ADC imaging (J0, J1). Lesion volumes and deep MRI embeddings are fused, reduced via PCA, and classified using a linear SVM.
• Figure 3: Validation AUC and F1 across tested models. J1 multimodal configurations outperformed J0 and clinical baselines.
• Figure 4: Feature importance derived from linear SVM coefficients
• Figure 5: MRI feature maps and histograms for two patients with different three-month mRS scores