Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

A Reliability-Aware Analytics Framework for Multi-Vendor Tuberculosis Screening from Chest Radiographs

Rektorat UDINUS 2026 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2026
Goals
Record
dc51e106-d527-482f-8fba-b5e0dac5b895
Published
2026-10-30

Universitas Dian Nuswantoro (UDINUS), with participation from Salam, Abu; Andono, Pulung Nurtantio; Purwanto; Soeleman, Moch Arief; Santoso, Antonius Gunawan; Dewi, Ika Novita; and Wright, Sasha-Gay, developed and evaluated TRUST-CXR, a reliability-aware workflow for tuberculosis screening from chest radiographs. The activity addressed challenges in real-world artificial intelligence deployment, including heterogeneous imaging systems, unreliable probability estimates, and limited operational evaluation of deployment-oriented screening workflows.

The study used a retrospective dataset of 3522 chest radiographs retrieved from a hospital Picture Archiving and Communication System in native Digital Imaging and Communications in Medicine format. The images covered FUJIFILM, SIEMENS, and PHILIPS imaging systems. The workflow integrated the PACS with automated triage and included image quality assessment, vendor-aware harmonization, artificial-intelligence-based tuberculosis risk prediction, probability calibration, subgroup-aware evaluation, and operational triage analytics.

Five workflow configurations were evaluated, comprising raw Digital Imaging and Communications in Medicine inference and four harmonization or preprocessing variants. Evaluation used a fixed triage threshold of 0.5. On a held-out test set of 200 examinations, TRUST-CXR achieved an F1-score of 0.869, sensitivity of 0.880, specificity of 0.853, an area under the receiver operating characteristic curve of 0.942, and an area under the precision–recall curve of 0.947. Post hoc calibration improved probability reliability. Operational analysis reduced the false-positive triage burden from 80.0 to 73.3 per 1000 examinations and reduced missed tuberculosis cases from 63.3 to 60.0 per 1000 examinations compared with the best baseline workflow, without materially increasing referral volume.

In relation to the stated goal, this UDINUS activity concerns health research focused on tuberculosis screening and deployment-oriented radiology workflows. Its institutional framing is the development and evaluation of a reliability-aware approach for multi-vendor chest-radiograph screening.

Reference https://www.scopus.com/pages/publications/105042295304?origin=resultslist