Member Connect

Log in

Login to your account

Username
Password *
Remember Me
Tuesday, 17 May 2016 13:12

Paper 2016_05

Rate this item
(0 votes)
An automated tuberculosis screening strategy combining X-ray-based computer-aided detection and clinical information

 

Jaime Melendez , Clara I. Sánchez , Rick H. H. M. Philipsen , Pragnya Maduskar , Rodney Dawson , Grant Theron, Keertan Dheda  & Bram van Ginneken

 

 

The study by Melendez et al. describes the use of Computer Aided Diagnosis (CAD) of chest X-rays for TB. CAD has been shown to improve the reading of chest X-rays and contribute to the diagnosis of TB. This is of great interest for settings with a lack of experienced X-ray readers and paves the way for a strategy of tele-radiology.

In the current study the authors show an improved performance of CAD when combined with a selection of 12 clinical features, of which some are assessed by the medical practioners and other are self-reported. The results are obtained by using machine-learning algorithms. This is compared with either CAD alone or clinical screening alone

 

Of interest is the finding that at a sensitivity of 95%, the combined strategy has a negative predictive value of 98%, a good tool to exclude TB or serve as a triage test for further investigations.

Intriguing is the finding that in the combined strategy typical symptoms like cough and chest pain do not play a role. This can be explained by the setting of the study (passive case finding with almost everybody coughing) and the population (South Africa with a high prevalence of HIV). This makes again very clear that diagnostic performance is driven by specific settings and that there is always a need for local validation of approaches.

 

(Comment written by Frank van Leth)

 

Full text here

Read 2353 times Last modified on Wednesday, 12 July 2017 17:56
More in this category: « Paper 2017_03 Paper 2016_11.1 »