Until now we have simply sent all the images of each M. tuberculosis sample growing on each and every of the 14 antibiotics out to be classified by the citizen scientists. A while ago we realised that some images are “easy” in the sense that all the volunteers we show it to all give exactly the same answer. So, with a bit of work behind the scenes, we’ve written some computer code that can detect where the wells are in each image and then measure the amount of growth in each well. Now, it is isn’t perfect; it can be confused by small amounts of growth and artefacts like air bubbles and shadows, but it does mean we can confidently filter out the relatively easy images, thereby allowing us to only send the more challenging cases to you, our volunteers.
Humans are much, much better than computers at interpreting images – you can instinctively tell the difference between an air bubble and something that might be growth. You can also compare what might be growth in a well to how the bug is growing in the control wells and decide that, whilst it might be growth, it is so much smaller that it isn’t worth bothering about.
The other driving force is we are starting to get a LOT of images, so this will bring the volume down to an achievable level, and hopefully allow BashTheBug to work through all of the more difficult cases our TB project is generating.
For more info, including some example images, please see this longer blog post on the Zooniverse website.