Computer vision is at the forefront of the machine learning revolution. Deep convolutional neural networks (CNNs), a class of machine learning algorithms, have transformed facial recognition, self-driving cars, and medical imaging. Neural networks were inspired by the way the brain is thought to process information and have now come full circle, aiding neuroscientists in studying the brain itself. In one recent example, scientists at North Carolina State University applied a deep CNN to study neurodegeneration in the roundworm.

The PVD neuron in the roundworm responds to physical touch and temperature, and has been found to undergo morphological changes as the animal ages. Specific regions become less organized and form protrusions or “beads” that correlate with age-related deficits in these sensations. Given the PVD neuron’s role in sensing temperature, these scientists were additionally interested in investigating whether there are similar morphological changes due to cold shock, a drastic reduction of temperature that can be deadly to these animals. However, monitoring the size, shape, and distributions of these beads from microscope images is a painstaking process, requiring scientists to spend hours or days manually circling each bead so the entire population can be analyzed. Enter deep learning.

These scientists trained a deep CNN to identify these beads with around 90% accuracy. Additional analysis allowed the group to study properties of the beads in various experimental conditions without manually identifying each one. For example, they found that older animals tended to have more beads and a smaller inter-bead distance (beads are closer together). In contrast, when the animals were exposed to cold temperatures (39 °F) for at least 16 hours, the bead number increased but the inter-bead distance decreased, indicating the beads tended to form in more remote regions of the neuron. Other tests examined the effect of allowing the animal to recover in warmer temperatures, wherein these hallmarks of neurodegeneration disappeared, depending on the severity of the cold shock.

Lastly, the authors leveraged another one of the benefits of machine learning algorithms: they often provide insight that may be hidden to scientists. By feeding the values for 46 different metrics about the bead distribution (such as average bead size, inter-bead distance, and bead number) into a classification algorithm, the computer could predict with over 80% accuracy whether the beads were from an old animal, a cold shocked animal, or neither, something a human likely couldn’t do.

Based on these results, scientists could ask if we can use CNNs to study neurodegeneration in our own brains. Can we identify patients in the early stages of Alzheimer’s disease or those suffering from mental illnesses? The potential is there, but many challenges must be overcome. Acquiring images from humans to analyze is more difficult, and we would need to apply the algorithms to the several billion neurons found in the human brain as well as improve the identification accuracy above 80-90%. Even so, as computers get more powerful and algorithms more advanced, we will likely see applications of machine learning in biology that we cannot even imagine yet.

Sahand Saberi-Bosari is a graduate of the San Miguel Lab with a PhD and is currently an employee of the BASF chemical company. Kevin Flores is an Assistant Professor and the Principal Investigator of the Flores laboratory for Mathematical Biology at NCSU. Adriana San Miguel is an Assistant Professor in the Department of Chemical and Biomolecular Engineering and the Principal Investigator of the San Miguel Lab at NC State University.

Managing Correspondent: Andrew T. Sullivan

Press Articles: How deep learning can advance study of neural degeneration,” ScienceDaily

New Research Shows How Deep Learning Can Help Advance Neural Degeneration Studies,” The Science Times

How AI Scratches More Than Just the Surface in Neurodegeneration,” Genetic Engineering & Biotechnology News

Original Journal Article: Deep learning-enabled analysis reveals distinct neuronal phenotypes induced by aging and cold-shock,” BMC Biology

Image Credit: Pixabay

Leave a Reply

Your email address will not be published.