Toward Markerless Video-based Tremor Analysis: Objective Quantification of Pathological Tremor in Mouse Preclinical Models
Yota Koshimoto, Akihiro Tsukahara, Yasuhiro Moriwaki, Mariko Isogawa, Mariko Isogawa
cs.CV
Quick verdict
RGB-only mouse tremor scoring reaches r=0.86 vs accelerometer PSD; well-engineered, but validated only on harmaline-challenged wild-type mice.
Summary
This paper presents a non-invasive tremor quantification pipeline for mice that uses only standard RGB cameras. From dual side-view videos of unrestrained mice on a suspension platform, the method segments the animal with SAM 2, tracks the head with DeepLabCut (ResNet-50), selects the higher-confidence view per 3-second segment, discards locomotion via a cumulative-displacement threshold (400 px ≈ 80 mm), and scores tremor by counting vertical head-oscillation peaks whose prominence exceeds 1 pixel (≈0.2 mm). On an original dataset of 56 synchronized front/back video sequences from 28 wild-type mice given harmaline (0, 5, 10, or 20 mg/kg), the tremor score correlates r=0.86 with accelerometer peak PSD, versus r=-0.27 for a prior velocity-based RGB baseline and r=0.82 for human expert scoring; ablations drop performance to 0.75-0.82, and hyperparameter sensitivity stays above 0.80. A one-way ANOVA (p<0.05) confirms the score captures dose-dependent tremor increases. Main limitations: only harmaline-challenged wild-type mice are studied (no disease models), the accelerometer ground truth is mounted on the platform rather than the animal, the unit of the correlation analysis is unspecified with no confidence intervals, no test distinguishes the method from human scoring, DeepLabCut's noise floor relative to the sub-millimeter threshold is unreported, and code/dataset release is not explicitly stated.
Review complete
Assessment at a glance
6
risk flags identified
● high 1
● medium 4
● low 1
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- A side-view RGB-only pipeline (SAM 2 segmentation + DeepLabCut head tracking + move detection + prominence-based peak counting) quantifies tremor severity with r=0.86 correlation to accelerometer peak PSD.
- The method outperforms a prior velocity-based RGB baseline (r=-0.27) and matches human expert scoring (r=0.82) while removing its manual time cost.
- Every proposed module contributes measurably, with ablation correlations of 0.75-0.82 when components are removed (Table 1).
- The estimated tremor score replicates harmaline's dose-dependent severity profile (one-way ANOVA p<0.05), supporting use in pharmacological evaluation.
- Side-view capture makes anti-gravity (vertical) tremor measurable, which top-down systems cannot observe.
Strengths
Full parameter transparency
Every operative constant is published — T=3 s segments, 400 px (~80 mm) move threshold, τ=1.0 px (~0.2 mm) prominence, 0.2 mm/px checkerboard calibration, DLC ResNet-50 trained on 480 annotated frames — enabling exact re-implementation by other labs.
Quantitative ground truth with ablations
An original 56-sequence synchronized video/accelerometer dataset supports the r=0.86 headline (Table 1), and ablations (0.75/0.78/0.82 without seg./Move Detection/multi-view) plus Fig. 2 sensitivity (>0.80 stable) substantiate each module's necessity rather than relying on qualitative claims.
Biological validation beyond correlation
Clear positional novelty
Noise-robust scoring design
Weaknesses
External validity gap
Indirect ground truth
Thin statistical reporting
Single-keypoint proxy without noise floor
Release not committed
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This cs.CV method paper (arXiv 2609.18753, Keio University) proposes a markerless pipeline that estimates mouse tremor severity from side-view RGB video alone. Four claims anchor it: (1) the tremor score correlates r=0.86 with accelerometer peak PSD (Table 1); (2) it beats the prior velocity-based RGB method of Ni et al. [16] (r=-0.27) and approaches human expert scoring (r=0.82) without its labor cost; (3) every module matters — removing segmentation, Move Detection, or multi-view selection drops r to 0.75, 0.78, or 0.82; (4) the score reproduces harmaline's dose-dependent tremor response (one-way ANOVA p<0.05; Figs. 3–4). Claimed novelty: capturing anti-gravity (vertical) tremor that top-down systems structurally miss.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 6 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
How can an ordinary RGB camera detect tremors too small to see?
The pipeline tracks the mouse's head with a pose-estimation model at 0.2 mm per pixel, then counts vertical oscillation peaks whose prominence exceeds 1 pixel (about 0.2 mm) over a 5-minute window. Individually sub-visual oscillations accumulate into a countable score, so severity can be ranked even when single peaks are hard to see.
How accurate is this video-based method compared to a gold standard?
On 28 mice given the tremor-inducing drug harmaline, the video-based score correlated r=0.86 with the peak power spectral density from an accelerometer, versus r=0.82 for a trained human scorer and r=-0.27 for a prior RGB-based method. It is roughly as accurate as expert visual scoring but requires no manual effort.
Was this tested on Parkinson's disease model mice?
No. The study used only wild-type (non-disease) mice challenged with harmaline, a drug that reliably induces tremor, at 0, 5, 10, and 20 mg/kg. Whether the method transfers to genetic or lesion-based Parkinson's models remains untested.
Is the code or dataset publicly available?
The paper lists a project page (isogawalab.github.io/Video-based-Tremor-Analysis-Project) but does not explicitly state that code, the trained model, or the video/accelerometer dataset will be released. All pipeline thresholds and setup parameters are, however, described in enough detail to re-implement.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 2-5 (sensitivity ranges, PSD and score distributions, tremor trace) cannot be verified from the extracted text; asterisk significance levels in Figs. 3-4 are unknown.
- The unit and sample size of the correlation analysis are not recoverable from the text.
- Code/dataset availability cannot be confirmed; only a project page URL is given.
- Mouse sex and strain details beyond 'wild-type' are absent from the extracted text.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.