What connects the Covid-19 vaccine, police forensics departments and the latest generation of computer processing chips? None of them would have been possible without microscopes.
Microscopy is responsible for many of the innovations that uphold modern living standards. Whether they’re using light, confocal, electron or X-ray microscopes, industries from pharma to software rely on these devices across their outputs.
So what new marvels might be in store if the process of using a microscope were itself to be revolutionised? We may soon find out. Today’s life scientists are harnessing automation and artificial intelligence (AI) technologies to radically streamline the microscopy process.
This can’t come soon enough. Microscopy’s end-to-end pipeline has long been rather onerous. This is true of all three steps: image acquisition, image analysis and the creation of useful outputs. But how are these technologies transforming workflows at each of these three stages? Let’s take a closer look.
Acquiring images
Acquiring a workable, high-quality image of the item under inspection requires manually calibrating a whole host of software and hardware parameters. This is both time- and labour-intensive and demands a lot of specialised domain knowledge.
To make matters more complicated, sometimes those parameters will need to be changed while image acquisition is underway. For instance, if the object under examination is growing or in any way altering its position under the lens, the settings will have to shift with it. This requires vigilance and agility from a human operator, creating a large potential for error and bias.
Another advantage of digitised microscopy is the image metadata it creates. The final file will contain information on the image, plus the microscope and camera settings that produced it. This provides valuable material for the analysis stage. It also makes experiments much easier to reproduce.
Analysing images
Once the microscope has done its work and produced a set of quality images, work can begin on finding out their secrets. Automation can then prepare the image for close analysis. Image enhancement software will remove noise and sharpen the definition for a crisp and clear picture. But the real AI magic happens at the segmentation stage.
Much medical research and clinical activity depends on detecting and classifying the contents of microscope images. For instance, when diagnosing cancer, researchers will need to be able to distinguish between normal and cancerous cells in a sample.
Done manually, this requires researchers to painstakingly inspect and annotate multiple datasets. But when AI is applied, everything changes. Researchers need only manually annotate a single dataset with the correct segments and then present it to a deep learning model for training. The model, now clued up on the taxonomy, can then automatically differentiate between segments in subsequent samples. This performs in seconds what could take researchers hours to finish. Multiply that across a stack of images and you have a huge scaling up of any lab’s productivity.
Creating outputs
The final stage of the microscopy pipeline is to turn the findings from analysis into useful outputs. Scientific reports require a great deal of very precise information. Creating these documents by hand can be almost as arduous as setting up their inputs in the first place. This is yet another step where automation comes in handy.
A report is normally accompanied by a log of all activities performed during the microscopy process, plus information about any errors. As you’d expect, this can contain a lot of technical details. But software used at the image acquisition and analysis stages will record all these factors as they occur and auto-generate a comprehensive metadata log.
With high-quality images also automatically stored on disk, ready for insertion throughout the report, many ingredients of a professional research output no longer require human labour.
Prioritise for success
AI could represent a step change in microscopy’s capabilities. But not every stage in every experiment will need to be automated from end to end. Sometimes, human input and oversight are very much necessary. Digital microscopes’ endless customisability gives researchers the freedom to find the optimal structure for every project.
It’s only a matter of time before an automated microscopy workflow helps a team stumble on life sciences’ next big discovery.