Abstract
Clinical gait assessment in neurological rehabilitation depends on the interpretation of specific movement features, yet most video-based approaches provide only diagnostic labels without structured explanations. We propose NeurGait, a vision-language framework for generating rubric-aligned gait annotations from video through knowledge distillation. A large vision-language model generates teacher annotations aligned with a structured gait rubric, which are then used to train a compact student model capable of producing model annotations together with diagnostic predictions. The data comprises gait videos from 77 participants (Parkinson’s Disease: N = 42, Hoehn & Yahr stages 1–3; Multiple Sclerosis: N = 35, EDSS scores 0–3), all clinically diagnosed by neurologists. We compared model annotations with independent blinded annotations from physiotherapists with at least ten years of neurorehabilitation experience, using the same structured gait rubric. Inter-rater agreement between physiotherapists for the binary PD/MS diagnosis task was only fair (κ = 0.349), highlighting the difficulty of distinguishing early-stage PD and MS from gait videos alone. Despite this challenge, agreement between model annotations and physiotherapist annotations for trunk lean, cadence, and walk speed reached moderate levels and approached physiotherapist agreement ranges. NeurGait-LUPI generated model annotations and diagnostic predictions directly from video at inference without requiring manual clinical labeling. The proposed framework achieved moderate agreement with physiotherapists on selected postural gait features, providing preliminary evidence that vision-language distillation can support rubric-aligned gait assessment from video.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMAI_037.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/profile?id=%7EGuzin_Kaya_Aytutuldu1
BibTex
@InProceedings{AytGuz_NeurGait_MICCAISAT2026,
author = { Aytutuldu, Guzin Kaya AND Aytutuldu, Ilhan AND Sakalli, Nazan Karagoz AND Celik, Rabia Gokcen Gozubatik AND Tutuncu, Mesude AND Ozdemir, Zeynep AND Huseyinsinoglu, Burcu Ersoz AND Akgul, Yusuf Sinan},
title = { { NeurGait: A Vision-Language Framework for Structured Gait Assessment Support in Parkinson’s Disease and Multiple Sclerosis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}