Recent publication shows how Aiforia® speeds up research on Parkinson’s disease.
A recent scientific study reveals that deep neural networks – one form of machine learning – applies well to neuron counting. When studying Parkinson’s disease, it is essential to recognize and count the nerve cells from microscopic brain sections. Stereology is the current gold standard to count the neurons, but it is very time-consuming and prone to human errors, since it requires a lot of manual hands-on work and interpretations can vary between the experts.
The researchers from the University of Helsinki compared the Aiforia® deep neural network algorithm with stereological method and found that the two methods yield highly similar results.
The study utilized brain tissue sections from rats and mice with experimentally induced Parkinson’s disease. Microscopic tissue sections of the brain were digitized with a whole-slide scanner and TH (tyrosine hydroxylase) -stained neurons were counted from substantia nigra both with stereology as well as the deep neural network algorithm. The results of neuronal cell count between the two methods were highly comparable; Pearson correlation of 0.9 or more with P-values less than 0.0001 was obtained both for the rat and mouse brain samples.
However, the deep neural network algorithm easily beats human observer in speed and endurance; it counted approximately 30 000 neurons in three hours, recognizing more than two neurons each second. Once the algorithm has been trained what to do, it continues tirelessly and can reproduce the calculation always when needed, eliminating the human error.
The deep neural network algorithms are a promising tool for the whole biomedical research community. Aiforia provides fast, accurate and high-throughput method for quantitative analysis of whole slide images. As a cloud-based platform, Aiforia is easily accessible for anyone, anywhere, through a web browser.