Performance of a novel algorithm using automated digital microscopy for diagnosing tuberculosis

Nazir A. Ismail, Shaheed V. Omar, James J. Lewis, David W. Dowdy, Andries W. Dreyer, Hermina Van Der Meulen, George Nconjana, David A. Clark, Gavin J. Churchyard

Research output: Contribution to journalArticlepeer-review


Rationale: TBDx automated microscopy is a novel technology that processes digital microscopic images to identify acid-fast bacilli (AFB). Use of TBDx as part of a diagnostic algorithm could improve the diagnosis of tuberculosis (TB), but its performance characteristics have not yet been formally tested. Objectives: To evaluate the performance of the TBDx automated microscopy system in algorithms for diagnosis of TB. Methods: Prospective samples from patients with presumed TB were processed in parallel with conventional smear microscopy, TBDx microscopy, and liquid culture. All TBDx-positive specimens were also tested with the Xpert MTB/RIF (GXP) assay. We evaluated the sensitivity and specificity of two algorithms-(1) TBDx-GXP (TBDx with positive specimens tested by Xpert MTB/RIF) and (2) TBDx alone-against the gold standard liquid media culture. Measurements and Main Results: Of 1,210 samples, 1,009 were eligible for evaluation, of which 109 were culture positive for Mycobacterium tuberculosis. The TBDx system identified 70 specimens (68 culture positive) as having 10 or more putative AFB (high positive) and 207 (19 culture positive) as having 1-9 putative AFB (low positive). An algorithm in which "low-positive" results on TBDx were confirmed by GXP had 78% sensitivity (85 of 109) and 99.8% specificity (889 of 900), requiring 21% (207 of 1,009) specimens to be processed by GXP. As a stand-alone test, a "high-positive" result on TBDx had 62% sensitivity and 99.7% specificity. Conclusions: TBDx used in diagnostic algorithms with GXP provided reasonable sensitivity and high specificity for active TB while dramatically reducing the number GXP tests performed. As a stand-alone microscopy system, its performance was equivalent to that of a highly experienced TB microscopist.

Original languageEnglish (US)
Pages (from-to)1443-1449
Number of pages7
JournalAmerican journal of respiratory and critical care medicine
Issue number12
StatePublished - Jun 15 2015


  • Microscopy
  • TBDx
  • Triage
  • Tuberculosis
  • Xpert

ASJC Scopus subject areas

  • Pulmonary and Respiratory Medicine
  • Critical Care and Intensive Care Medicine


Dive into the research topics of 'Performance of a novel algorithm using automated digital microscopy for diagnosing tuberculosis'. Together they form a unique fingerprint.

Cite this