Study Rationale:
Recent advances in artificial intelligence and, in particular, in one of its methods called deep learning offer exciting opportunities for the diagnosis and treatment of Parkinson’s disease (PD). Using convolutional neural networks, a deep learning technique, we were able to discover information in microscopic images that humans can’t comprehend. We think these techniques could be used to gain insight from microscopic images of tissue samples donated by people with PD.
Hypothesis:
This study aims to determine whether a deep learning network can be trained to detect known Parkinson’s-associated changes as well as or better than a trained pathologist and do so faster, cheaper and more consistently. The other aim of this study is to determine whether a deep learning network can be trained to discover differences in tissue samples from people with Parkinson’s disease that are not obvious to pathologists. This would allow predictions of disease stage and progression earlier than can be done currently.
Study Design:
We propose to train deep learning convolutional neural networks on samples from people with PD in an effort to develop an algorithm that could diagnose and stage the disease more effectively and reliably than is currently possible.
Impact on Diagnosis/Treatment of Parkinson’s Disease:
If successful, the algorithm could be disseminated and scaled, potentially providing an earlier diagnosis and a more quantitative, uniform and inexpensive tool for assessment of samples in large clinical studies.
Next Steps for Development:
At the end of this study, we will have a good understanding of the quality of the predictions based on the data we currently have available. It would be important to first test the developed algorithms on a completely independent data set to be confident that the technology is generalizable and ready for widespread use.