Skip to content
TAKOTSUBO®Life Sciences
Let's talk
Data Science & Analytics

How Data Science Structured a Review of Infrared Thermography and Occupational Health

4 min read

Editorial illustration of the path from bibliographic data to auditable evidence in thermography and occupational health

Published research | Exploration of Musculoskeletal Diseases, 2026

From a clinical question to auditable data: how data science structured a review of thermography and work

A review of infrared thermography in occupational health does not begin with a thermal image. It begins by turning a heterogeneous literature into a body of evidence that is traceable, testable and useful for responsible decisions.

In Standardizing infrared thermography for occupational applications, published in Exploration of Musculoskeletal Diseases, the Research and Development Division – Data Science & Analytics at Takotsubo Life Sciences contributed to the study’s methodological core: search design, record curation, quality control, formal analysis and construction of a reproducible evidence trail.

The final product is an integrative narrative review of work-related musculoskeletal disorders. Its value, however, extends beyond clinical synthesis: it shows how to organise complex bibliographic data without hiding uncertainty, double-counting studies or turning thermal signals into unjustified diagnostic claims.

PRISMA flow diagram for identification, screening, eligibility and inclusion of studies in the review
Evidence has a journey too. The diagram shows how records move through identification, deduplication, screening and eligibility before forming the final corpus. Source: Ribeiro & Giacomini (2026), Figure 1, adapted from PRISMA 2020 (CC BY).
1,128records identified
66duplicates removed deterministically
247studies in the final corpus
5complementary bibliographic databases

Search is the first layer of the data model

To answer the question of how thermography may support screening and monitoring of occupational overload, the search was structured around four conceptual blocks: thermography, musculoskeletal conditions, work/ergonomics context, and screening or monitoring. The strategies were adapted to the syntax of PubMed/MEDLINE, Embase, Scopus, Web of Science and SciELO, while the complete queries were retained as supplementary material.

This stage can look merely bibliographic, but it determines every downstream analysis: vocabulary, search fields, filters and language criteria define the universe that can be examined. The study also combined database searches and citation tracking, preserving the source of each entry.

1. Preserve provenanceExported files were retained as raw inputs, with metadata and an SHA-256 manifest. Each record received source fields: database, file and import position. Every decision can therefore be linked back to its starting data.
2. Normalise before comparingDOI, PMID, title, year and journal were standardised. Normalising URLs, case, whitespace and formats reduces writing noise without merging distinct records.
3. Deduplicate with explicit rulesThe hierarchy DOI > PMID > normalised title + year > normalised title organised duplicate clusters. Within each cluster, a fixed rule prioritised bibliographic completeness and, in a tie, a pre-specified source. Nothing is removed opaquely: each discarded item is mapped to the record retained.
4. Turn screening into auditable evidenceTwo reviewers independently assessed title and abstract. Disagreements were flagged, conservatively retained until adjudication, and prevented the workflow from closing while unresolved. The system turns human disagreement into a verifiable event rather than a lost note.

What changes in practice

Structured data help constrain conclusions, not merely expand them

Thermography as a complementary functional layer

The synthesis supports value in relative patterns—bilateral asymmetry, task-induced change and recovery—when acquisition is controlled. It does not support replacing clinical diagnosis or structural imaging with thermography.

Protocol is part of the result

Ambient temperature, humidity, airflow, acclimatisation, emissivity, acquisition geometry and ROI definition determine whether a thermal signal is comparable. Without those variables, a temperature difference may simply be noise.

Multimodal integration is an operational hypothesis

The proposed pathway combines thermography with EMG, dynamometry, ultrasonography and worker-reported outcomes. It is promising for monitoring, but still requires longitudinal, multicentre and cross-device validation.

Where Data Science & Analytics becomes indispensable

The article does not use “data science” as a label for automating a review. It applies it as a quality discipline: immutable inputs, standardised decisions, deterministic rules, missingness codes that distinguish “not reported” from “not applicable”, logical checks and audit artefacts.

That design matters especially in thermography. Because devices, ROIs, protocols and metrics differ considerably between studies, responsible synthesis must preserve heterogeneity instead of manufacturing precision. That is why the work used thematic integration rather than a single-effect meta-analysis.

Reproducibility is a deliverable, not a promise

The scripts and analytical structure were made public. Other groups can inspect search logic, deduplication, decision tables and generation of PRISMA counts—without redistributing commercial database exports that are protected by licence.

This is an important balance: methodological transparency sufficient for audit, alongside respect for the permitted use of source data.

Science in every decision

The better question is not whether a thermal image “diagnoses”

The more useful question is how to build protocols, comparable data and traceable decisions so that a functional signal can be interpreted alongside clinical assessment and other measures. That is where data science turns a review into evidence infrastructure.

Read the published article · Access scripts and reproducibility materials