Tecniplast DVC® · Analysis pipelines · 2026-06-25

Inside the DVC analysis pipeline — the methods behind the metrics

A list of papers tells you what was measured. This deck tells you how: one raw signal forking into a connected family of biomarkers — each with its recipe, its thresholds, and the cage configurations it is valid for — and why the real constraint on reuse is metadata, not the maths.

Damien Huzard, PhD · Neuronautix

How this was built — a second brain at work

From a vendor paper list to a methods map

Starting from the Tecniplast DVC® scientific-papers catalogue, a small team of AI research assistants read the open-access full texts in parallel — each taking a few papers, each pulling out exactly how the data were processed. Findings were checked against the PDFs and written into the Neuronautix knowledge base, which the assistants read before writing and add to after. The knowledge compounds: a paper list became a living, cited methods resource.

Source: digitalcage-tecniplast.com/en/scientific-papers.html · extracted from the open-access methods sections

The raw signal — where every metric begins

Twelve electrodes, four times a second

Electrodes / cage

12

A 4×3 capacitive grid under each cage floor — no camera, no implant [1].

Sampling rate

4 Hz

Every electrode read every 250 ms; an "activation" is a thresholded change [1].

Data / cage / day

~2.5 MB

Every named metric is a transformation of this twelve-channel series [1, 2].

Iannello, Heliyon 2019 · Pernold et al., PLoS One 2023 [1, 2]

Activity · Two algorithms — and the cage rule

Activation (any cage) vs tracking (single-housed)

Activation / ALI — valid for all cage types & numerosities Two-window capacitance difference thresholded at λ = 1.25 (empty-cage noise floor) → binary per electrode, averaged per second. Cage-level, so it travels everywhere [2].

Centroid tracking — single-housed only Signal-drop-weighted centroid (~1 mm); motion ≥ 1 mm/sample; 65 mm stride splits movement-on-the-spot from locomotion; long rest ≥ 40 s. Needs one animal per cage [2].

The rule that cuts across Amount/rhythm metrics (ALI, RDI, circadian, region-spatial) work group-housed; only trajectory metrics (distance, bouts) require single housing [2].

Pernold et al., PLoS One 2023 — the fully specified recipes [2]

Biomarker · Rest Disturbance Index (RDI)

A precise algorithm, not a vibe

The recipe (one origin)

Minute-bin the activity series

Zero everything below λ = 0.005

4th-order Butterworth band-pass (1/2000, 1/300)

Sample entropy, m = 2, r = 0.2

Why it matters

Measures rest fragmentation, not rest amount

Scale-invariant; self-referential baseline

Defined once; every later paper reuses it

Golini et al., Front Neurosci 2020 — the RDI definition [3]

Biomarker · Circadian / regularity

Period, stability, amplitude

The circadian chain is standard chronobiology, fully reimplementable: a 30-minute running average, then a chi-square periodogram in the ImageJ plug-in ActogramJ, with inter-daily stability (IS), intra-daily variability (IV), and relative amplitude (RA) hand-coded in R from the Witting equations — all under a per-cage LED system at ~100 photopic lux, so each cage is its own light-controlled chamber [4].

Tir et al., Sci Rep 2025 — circadian phenotyping at rack scale [4]

Biomarker · Spatial preference

Read straight off the grid

Frontality

Share of activity over the six frontal electrodes (7–12) [5].

Wall activity

Sum of the left and right electrode columns — a thigmotaxis proxy [5].

Gini index

Concentration of activity across the twelve electrodes (0 = even, 1 = one spot) [5].

Group-housed-friendly, and derived from previously recorded data with no new animals — data repurposing for the Reduction principle.

Fuochi, Rigamonti et al., Sci Rep 2023 [5]

Scaling to real colonies · A contested step

Per-animal = cage ÷ mice — with a caveat

For group-housed cages, a per-animal signal is often obtained by dividing the hourly cage activity index by the number of mice (each cage single-genotype), validated against a TSE LabMaster beam frame via z-scores. But this is not universally accepted: a cage aggregate cannot cleanly be attributed to one animal — activity is super-additive when mice interact, sub-additive when they huddle — so dividing by occupancy is a defensible average, not a per-individual measurement, and should be justified per design rather than assumed [6].

Sun et al., Front Neurosci 2024 [6] · normalisation caveat: Neuronautix

Environmental channels · From bedding to biology

The Urination Index is deterministic

The transform (no ML)

Bedding moisture → drop in EM field → Bedding Status Index

Invert each moisture increment

Exclude deltas around cage-insertion events

Cumulative sum ÷ housing density

Calibration & payoff

~1.4 index units per mL of water

Tracks blood glucose; flags hyperglycemia early

Code is public — UrinatoR (R/Shiny, MIT)

Brachs et al., Lab Anim (NY) 2025 · BSI cage-change ML: Collins, JAALAS 2025 [7, 8]

Machine learning on DVC signals

The transparent classifier worth copying

Colitis severity — logistic regression Two features only — DVC activity + body weight — trained on a single day (day 7), threshold 0.5. Simple, legible, reimplementable from the paper [9].

Cage-change — supervised ML Trained on human annotations of soiled bedding; >90% accuracy at high density; supports 3–6-week intervals. Model internals stay behind the paywall [8].

Narcolepsy — nest algorithm An inability to sustain activity for > 40 minutes is itself a robust biomarker, with a nest-identification step [10].

Zentrich 2021 · Collins 2025 · Piilgaard 2023 [8–10]

The finding that matters most

The real gap is metadata, not maths

Raw data: open in principle, unusable in practice

The raw electrode stream is fully exportable — yet rarely usable: it sits in per-facility silos and, decisively, arrives without the metadata (strain, sex, age, density, cage-change schedule, light programme, intervention timestamps) that every recipe quietly depends on. Compounding it, the analysis code is unreleased — UrinatoR (open-source, MIT) aside — and the first transform runs inside proprietary DVC Analytics. So the binding constraint on reuse is data + metadata, not algorithms: a reproducible recipe is worthless without well-described ingredients — exactly the FAIR-metadata, standard-format, API discipline recent reviews call for.

UrinatoR: github.com/Mortendall/UrinatoR · standards & metadata call: Fuochi et al., Front Big Data 2024 [7, 11, 12]

For DVC users · What to do with this

Treat every metric as provenance

Record recipe + context Store, with each value, the derivation, the baseline, the cage configuration, and the software version — the metadata that decides whether it is reusable [11].

Reuse the published thresholds EAD λ = 1.25 · long rest ≥ 40 s · RDI entropy (m = 2, r = 0.2) · frontality = electrodes 7–12 — but check the cage rule first [2–5].

The second-brain payoff A curated knowledge base turned a vendor paper list into a cited methods resource — and kept it, for the next question [12].

A FAIR, reusable metric stack — known before it is analysed

References — full citations with DOIs

Sources

[1]Iannello F. Non-intrusive high-throughput automated data collection from the home cage. Heliyon 2019;5(4):e01454. doi.org/10.1016/j.heliyon.2019.e01454

[2]Pernold K. et al. Bouts of rest and physical activity in C57BL/6J mice. PLoS One 2023;18(1):e0280416. doi.org/10.1371/journal.pone.0280416

[3]Golini E. et al. A non-invasive digital biomarker for rest disturbances in the SOD1^G93A ALS model. Front Neurosci 2020;14:896. doi.org/10.3389/fnins.2020.00896

[4]Tir S. et al. Evaluation of the DVC® system for circadian phenotyping. Sci Rep 2025;15:s41598-025-87530-6. doi.org/10.1038/s41598-025-87530-6

[5]Fuochi S., Rigamonti M. et al. Data repurposing from digital home-cage monitoring. Sci Rep 2023;13:10851. doi.org/10.1038/s41598-023-37464-8

[6]Sun R. et al. Accurate locomotor activity profiles of group-housed mice. Front Neurosci 2024;18:1456307. doi.org/10.3389/fnins.2024.1456307

[7]Brachs S. et al. Robust non-invasive detection of hyperglycemia using the Urination Index. Lab Anim (NY) 2025;54(12):379–389. doi.org/10.1038/s41684-025-01648-8

[8]Collins J.M. et al. ML/AI to determine cage-change frequency (Bedding Status Index). JAALAS 2025;64(4). doi.org/10.30802/AALAS-JAALAS-24-151

[9]Zentrich E. et al. Automated home-cage monitoring during acute experimental colitis. Front Neurosci 2021;15:760606. doi.org/10.3389/fnins.2021.760606

[10]Piilgaard L. et al. Non-invasive detection of narcolepsy type I (HCRT-KO, DTA). Sleep 2023;46(11):zsad144. doi.org/10.1093/sleep/zsad144

[11]Fuochi S. et al. Big data and its impact on the 3Rs: a home-cage monitoring review. Front Big Data 2024;7:1390467. doi.org/10.3389/fdata.2024.1390467

[12]Dall M., Brachs S. UrinatoR — open-source R/Shiny app for the Urination Index (MIT). github.com/Mortendall/UrinatoR

Thanks.

One signal, a connected family of biomarkers — know how each number was computed and for which cage setup it is valid, then store the recipe and the metadata with the result. The maths reproduces; the metadata is what makes it reusable. Built from the published methods sections, curated in the Neuronautix knowledge base.

Damien Huzard, PhD · Neuronautix · 2026-06-25
neuronautix.com/contact