🏠 Overview
impress-stroke.com β†—
🏠How does stroke lesion location relate to motor outcomes at 12 weeks?
Explore segmentation results and outcome predictions from 177 stroke patients. Compare 8 deep learning loss functions β€” and see how lesion location in motor pathways relates to upper-limb and walking outcomes at 12 weeks.
Research tool β€” not for deployment. This dashboard accompanies a journal manuscript (in preparation) from the University of Auckland. All results are from the UOA IMPRESS cohort (N=177 MRI scans, N=149 matched patients). Use the guided tour below to explore the results, or jump straight to a section.
πŸ”₯Demo Patient
Example stroke case Β· real MRI data
πŸ₯Demo Patient
Predicted FM-UE at 12 Weeks:
38/ 66range 32–44
Walking Outcome: Borderline Independent
Moderate CST damage is associated with limited upper-limb motor improvement at 12 weeks.
πŸ” Why this prediction?
🧠CST overlap:34%
🎯Motor cortex (M1):18%
πŸ“Lesion size:Medium
πŸ“ŠSegmentation DSC:0.741
Age: 68 yearsNIHSS: 8SAFE: 52/100FM-UE baseline: 28/66Lesion volume: 12,400 voxels
axial MRI
Lesion (GT)
CST overlap
Prediction
Scroll to zoom Β· drag to pan Β· β›Ά fullscreen Β· toggle overlay
πŸ§ͺExplore Step-by-Step
How to Use This Dashboard
A guide for first-time visitors
🎯What is this dashboard?Step 1 of 4

This is an interactive research companion to a journal manuscript (in preparation) on stroke lesion segmentation and outcome prediction. It lets you explore all results interactively β€” rather than reading static tables in a paper.

The study uses the UOA IMPRESS cohort (177 MRI scans, 149 matched patients) to compare 9 deep learning loss functions for lesion segmentation, then uses the segmentation outputs to predict upper-limb motor function (FM-UE) and walking ability (FAC) at 12 weeks post-stroke.

Research Pipeline
From multi-centre datasets to clinical outcome prediction
ISLES 2018
94 CT & CTPs
UOA IMPRESS
177 MRI
ATLAS 2.0
655 MRI
↓
πŸ”¬ Lesion Segmentation
↓
Contribution 1
New Regional Loss Function
Compared with 7 other losses Β· Tiny lesion: Ours & GDice best
Contribution 2
Channel Attention Block
Added to UNet, UNet++, Attention UNet Β· Ensemble: ATLAS 72.3 DSC
↓
πŸ—ΊοΈ JHU CST + SMATT Motor Atlas Overlap Features
UOA: 8 loss segmentations Β· Motor features + Clinical features
↓
🦾 Upper-Limb
FM-UE at 12 weeks
N=149 matched patients Β· Threshold 50
🚢 Walking Ability
FAC at 12 weeks
N=57 / N=46 Β· FAC β‰₯ 4
↓
Clinical Features Dominated β€” Imaging adds genuine value when baseline function unknown
IMPRESS Cohort Summary
Descriptive statistics for the UOA IMPRESS stroke cohort
Loading...
Explore the Dashboard
Click any section to navigate