Diagnostics 229
Hours that the Medical College of Wisconsin could save on testing urine samples every year by using digital imaging technology and AI to autoreport
negative specimens. Journal of Clinical Microbiology
the implementation. Using automation to move from one-shift to three-shift plate readings for urine cut a day from the clinical workflow by ensuring results reached doctors in time for the early morning slot – when they review lab reports – rather than in the early afternoon. The drop in manual interventions also meant that cultures could be incubated under far more consistent conditions, which, when combined with high-resolution imaging, meant the lab was able to read urine plates after 16 hours rather than 24.
The impacts on surveillance and susceptibility testing more generally were just as impressive. “We really started to see better isolation of organisms,” says Ledeboer. “Our colonies were bigger, so we could do susceptibilities faster. Instead of having to hold the plate for 48 hours, we were seeing [answers] on the first read of the plate – again, because of much better imaging capabilities. We really had a turnaround benefit, a susceptibility turnaround benefit, and we had an improved isolation benefit that was attached to it.” Whether it means finding an infection, preventing transmission or identifying a treatment faster, all of these improvements have direct impacts on patient outcomes.
“If we can just remove the variation we see between your eyes and my eyes when we both look at a slide, we can be much more consistent at screening for rare events.”
Nate Ledeboer
Now, with so much evidence to justify the initial capital outlay and change in behaviour, Ledeboer notes how it’s much easier for a laboratory to “jump feet first” into automation and update all of its workflows at once. Nonetheless, Lenk, for all his enthusiasm, is careful to break the process of digitally transforming a lab into five tiers. The first, which simply requires the use of devices that deliver data through sensors, is ubiquitous. In the fifth tier, all of those sensors are networked together and feed into a data lake that is structured, contextualised and interpreted for technicians by machine learning and AI tools. Labs that can do that are vanishingly rare. But Lenk doesn’t blame laboratories – which he feels are generally stuck in tier three, feeding data into a unified storage device but struggling to make use of it thereafter – for failing to reach that point. Device and software developers need to work out how to standardise their offerings too. Without shared interfaces for data transfer, digitally transformed lab staff will have to do as much to manage virtual samples as they once did for physical ones.
32
“The typical business of a lab supplier is just delivering the devices,” he explains. “They say, ‘Okay, you are the researcher, do what you want.’ This gap needs to also be filled right now because all the devices are getting so complicated and specific, [and] somebody has to advise the researchers how to connect these devices.” As his research group at TU Dresden is a hub for smart lab research and implementation in Europe, Lenk has had as much work convincing developers to adapt as Ledeboer has with technologists. “Most of the vendors are small and medium-sized enterprises,” he continues. “So, you’re usually talking to the CEO and founder of the company, who says, ‘We have been doing this for the last 20 years – why do we need to adapt now?’ Not everybody takes these discussions light-heartedly.” Previously, Lenk had to struggle to get developers to adapt to both OPC Unified Architecture and standardisation in lab automation (SiLA) consortium standards. Over the past year, the two have cohered into an ecosystem whereby the first is used for devices that communicate primarily with other devices, and the latter for devices that interact more with users, making things considerably easier.
Measurable results
Since implementing its automation system, Ledeboer’s research team has focused on using built-in digital imaging capabilities and AI algorithms to achieve some of the loftier goals of digital transformation – even turning the job of plate reading over to its machines. His recent Journal of Clinical Microbiology paper on using segregation software to automatically analyse urine cultures in commonly-used 5% sheep’s blood and MacConkey agars found it had an overall sensitivity of 99.8% and a specificity of 72% – making it highly reliable for batch or autoreporting negative specimens. With 43.3% of all specimens registering as negative to both the algorithm and technologists, that could mean 55,000 fewer urine cultures in need of individual technologist review every year in Ledeboer’s lab – a saving of approximately 229 hours. Although Ledeboer’s study did not explicitly address variations in how different technologists interpreted cultures, there’s a growing body of evidence suggesting that machine-learning tools, which don’t get tired or distracted, can be more consistent than lab staff. It’s a possibility Ledeboer intends to pursue. “If we can just remove the variation we see between your eyes and my eyes when we both look at a slide, we can be much more consistent at screening for rare events,” he says. “There’s a huge opportunity to improve quality of care and improve consistency in how we report things.” That said, regulators don’t have to approve whatever it is about the human mind that might make us see things differently. They will have to develop a unified understanding of machine learning – which will not leave our brains the way it found them.
Practical Patient Care /
www.practical-patient-care.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59
