• There are no suggestions because the search field is empty.

Barrow Neurological Institute case study: stereology versus AI in neuron quantification

Stereology has long been a reliable method for neuron quantification, but how does it compare in speed and accuracy to artificial intelligence based tools?
Written by Aiforia
 

Neuroscientists Fredric Manfredsson and Ivette Sandoval, both currently at the Barrow Neurological Institute, had never considered using deep learning AI. Their research on gene therapies for Parkinson’s disease felt far connected to artificial intelligence.

However, after meeting an Aiforia scientist at a conference, they were intrigued by the potential of this software as a tool to challenge current methods of neuron quantification with stereology, an extremely time-consuming part of their lab’s work. This alternative tool was met with some hesitation at first, as Fredric mentions: “I was skeptical that a computer or software could figure out this analysis.”

Comparing AI to stereology

The two scientists embarked upon their first experimentation with deep learning AI. “The Aiforia methods and stats were impressive. The results produced were comparable to what we would get with our traditional methods using stereology,” Fredric describes. “When the best technician in the lab did this analysis, compared to Aiforia, the relative difference in results produced was very small.” Aiforia, therefore, provided promising, comparable results, assuaging the scientists’ skepticism.

What impressed Fredric and Ivette even more was the speed at which this analysis was conducted: “With a computer technician interface, it takes a couple of hours per animal.” Therefore, with the 3,000 sections analyzed as part of this scope of work, Ivette explains:

“It would have taken 20 workdays with stereology. Aiforia did this in a few days.” 


When asked how they both felt about deep learning AI now, both replied: “We turned from skeptics to believers.”

Automating manual image analysis tasks

Automating this once time-consuming analysis grants scientists and the rest of their lab significantly more time to focus on complex tasks. Hours were spent training PhD students, who would then spend many hours counting these cells. As a PhD student left the lab, another one would have to be trained.

Subjectivity can arise between counters, resulting in the potential for further issues. “The field is set on traditional ways of counting; it takes up so much time. I was excited the Aiforia software was validated,” Fredric adds. Both neuroscientists, now self-proclaimed believers, are excited to explore other ways in which deep learning can enhance their research.

Read more neuro case studies: