Calibrating computer-aided detection (CAD) for TB

Calibrating computer-aided detection (CAD) for TB

Overview

A research toolkit to support the effective use of computer-aided detection (CAD) software for TB by calibrating CAD score thresholds and other parameters. 

Chest radiography (CXR) plays a key role in the screening and triage of pulmonary tuberculosis (TB) and can guide the effective use of diagnostic testing to improve case detection and cost-efficiency.  Computer-aided detection (CAD) products use artificial intelligence (AI) to analyse CXR for the presence of abnormalities suggestive of pulmonary TB and can improve the feasibility and performance of CXR for TB screening and triage.  CAD technologies for TB detection have recently been recommended for use by WHO among adults aged 15 years or more, in place of human readers for interpretation of digital chest radiography in both screening and triage for TB disease.

CAD products produce an abnormality score that can be used to signal probable TB and trigger further TB diagnostic evaluation relative to a selected threshold. Identifying an appropriate threshold requires the calibration of CAD products based on the local context and intended use, as well as decision making around the goals for screening and acceptable costs.

Purpose

This toolkit has been developed to support calibration studies to facilitate the implementation of CAD for TB screening and detection. The toolkit has been designed for national TB programmes (NTPs) and other implementers who have decided to implement CAD.  

The specific objectives of this toolkit are to:  

  • Support new users of CAD to understand threshold scores and their programmatic implications within the context of a TB screening or triage programme.
  • Describe a simplified operational research protocol to collect and analyse the required data necessary to determine the appropriate CAD threshold for a specific CAD implementation.
  • Support the analysis, interpretation and application of the resulting CAD diagnostic performance data and cost estimates to determine the most appropriate CAD thresholds based on local context and use case.


By providing a standardised methodology, this toolkit will support the collection and analysis of comparable data on CAD performance across various settings that can be used to better inform the use of CAD globally. 

For further information, please contact Dr Corinne Merle at merlec@who.int



Factsheet