The evaluation in South Africa is intended to quantify the performance comparisons between TBDx™ and routine microscopy, and assess how automated sputum microscopy can be combined with other diagnostic technologies not only to deliver improved performance results but also to offer important cost-related savings. This analysis will be available soon. In the meantime we would like to share important observations gleaned from the evaluation.
During our time with the National TB Reference Laboratory (NTBRL) we were able to work closely with the laboratory staff to better understand their workflow. We loaded and read 1,249 cases from TB suspects and we tested new target slides and multiple detection algorithms. We discussed the technology with the staff and gained additional operational insights, a sample of which we note below:
- Target slides, having a designated circle in the middle, will become an operational necessity.
The designated location results in a more dependably applied smear and a more consistent and higher quality of staining. The enhanced quality of the stained smear produces better focused images, and that contributes to better detection performance.
- Slide quality matters greatly.
Quality slides produce smoother operations of the slide loader and automated stage, and lead to better performance results. Lower quality slides, often imprecisely cut, can lead to loader or automated stage malfunctions.
- The laboratory staff desires diagnostic results as soon as possible.
Though the software application has been designed to inventory images, thereby providing the technician an opportunity to review images from scanty cases, there is no need for this review when TBDx™ positive cases are confirmed by PCR or culture. Load the slides. Analyze the Images. Detect MTB. Print a Report. Keep it Simple!
- Warehousing digital images on a storage device may be unnecessary.
Maintaining images of cases determined to be normal is unneeded. Keeping the one image from positive cases containing the highest count of MTB would be sufficient. Create a 30-day purge routine. Keep it Simple!
In addition to the lessons learned we have made several observations that we believe will be confirmed when the final results are made available:
- We tested two different detection algorithms against each stained smear. It is our belief that one was a clear winner, producing what we believe will be a strong balance in both sensitivity and specificity performance. New classifiers developed for this algorithm, and designed to remove likely ‘false positive’ objects, seemed from the data available to have performed very well.
- We also tested the detection performance against the capture and analysis of a variety of field of view (FOV) captures – 100, 200, and 300. When the final results are known we may learn that acquiring 200 or 300 FOVs has very little impact on sensitivity and specificity performances. Acquiring fewer FOVs could add operational efficiencies and faster diagnostic results.
- The upcoming culture results will help determine what TBDx™ positive cases should be confirmed and tested for drug susceptibility by secondary diagnostic applications. Soon it will be better understood whether scanty cases containing only 1 or 2 bacilli will deliver the detection benefits to offset the costs of confirmation by secondary procedures.
In a world where TB remains a constant healthcare concern and major economic drag on developing nations’ economies, a low-cost and high-performing diagnostic application will play a significant diagnostic role. An automated sputum microscopy platform like TBDx™ will play an exciting role in supplementing human vision and improving diagnostic performance. We look forward to sharing the analysis as soon as it is permissible.