§ Science

Grounded in the literature. Validated in the clinic.

The Telephos measurement engine builds on a decade of peer-reviewed work in digital biomarkers, dyadic facial-emotion dynamics, and computational psychiatry — translated into a deployable platform.

§ 01 — Methodological lineage

What we're building on.

PNAS · 2024

Automating the analysis of facial emotion expression dynamics: A computational framework and application in psychotic disorders

Hall NT, Hallquist MN, Martin EA, Lian W, Jonas KG, Kotov R. Proc Natl Acad Sci U S A. 121(14):e2313665121.

Introduces a machine-learning + network-modelling pipeline for quantifying brief facial emotion expression dynamics from clinical interview video. Validated on 96 patients with psychotic disorders and 116 controls, the approach combined automated FER, ARMA pre-whitening, and VAR(1) network modelling to recover diagnostically meaningful patterns — schizophrenia trajectories drifting toward uncommon expressions (fear, surprise), other psychoses toward sadness.

npj Digital Medicine · 2021

Using a Digital Neuro Signature to measure longitudinal individual-level change in Alzheimer's disease: the Altoida large cohort study

Meier IB, Buegler M, Harms R, Seixas A, Çöltekin A, Tarnanas I (senior/corresponding author — now Telephos's CSO, then at Altoida Inc.). npj Digit Med. 4:101. N = 525 participants across two semi-naturalistic experiments.

Validates a 10-minute active digital cognitive assessment (Altoida DNS) against conventional 45–120-minute neuropsychological batteries across HC, MCI, and AD groups. The methodological contribution Telephos draws on is the use of intraindividual variability (IIV) — and its longitudinal form (LIIV) — as a dispersion-based sensitivity marker, shown to be more sensitive than conventional NP for detecting disease-trajectory change, especially at pre-conversion events.

§ 02 — Proprietary IP

Two pending patents. A defensible architecture.

Telephos has filed two patents covering the proprietary architecture behind AI-COPE — Time-Series Large Video Models and a Digital Twin Patch Encoder for direct temporal reasoning over multimodal patient–clinician data.

GB 2511888.6

Time-Series Large Video Models for clinical interaction analysis

GB 2514969.1

Digital Twin Patch Encoder — multimodal temporal reasoning

Technical brief — AI-COPE

Read the full mathematical and clinical detail.

Eight-page PDF: TSLVM architecture, the LPRS construction, the ANCOVA integration, and the validation results. Written for sponsors, biostatisticians, and life-sciences investors.

REVIEW Download — coming soon
Free, ungated. PDF, ~1.4MB.

Want to dig deeper into the methods?

We're happy to walk research partners and biostatisticians through the full pipeline, including the unpublished validation work.

Request a technical session ↗