STTR Phase II: Deep Learning Technology For The Microscopic Analysis Of Stained Cells Using Unbiased Methods

Award ID: 1926990

The broader impact/commercial potential of this Small Business Technology Transfer (STTR) Phase II project is the development of software to help bioscientists analyze more tissue in less time and with higher accuracy and reproducibility. Currently, stereology studies require a trained technician to sit before a computer screen, making tedious manual counts (clicks) on hundreds to thousands of microscopic cells. The technology under development uses a series of algorithms that combine unbiased stereology with a powerful new approach - deep learning (artificial intelligence) - to quantify stereology changes in stained cells in an objective, faster and more reproducible manner than existing methods. A major advantage of automating stereology using an objective deep learning approach is the savings in time and costs currently required of a trained technician to collect data. This technology will improve the pace of research into the causes of human diseases and accelerate the development of novel strategies for the therapeutic management of afflicted patients. Bioscientists and medical scientists will better serve the needs of society for a greater understanding of cellular structure in health, aging and disease, while the technology will ensure accurate and precise stereology results without the need for highly trained data collectors and at a small fraction of the time and cost of existing approaches. This STTR Phase II project proposes to develop software to automate the collection of stereology data in biomedical and bioscience research. The results of the Phase I research showed that tissue sections were automatically counted over 10X times faster with equal accuracy as manual stereology. In addition, reproducibility was 99% and required little or no user training. The Phase II objectives are to develop standardized, high-throughput, deep learning networks for quantifying other stereology parameters of stained cells and tissues, validate active deep learning as a technique for customizing deep learning for all user staining protocols, and update the software and documentation to support user-friendly workflow. The aim is to train the neural network to accurately segment a wider range of cell and tissue pathology with variable morphologies with a performance metric for accuracy of 95%. Unbiased stereology will allow bioscientists to accurately quantify the cellular changes that cause metabolic diseases, neurological disorders, toxic reactions and mental illnesses. Automating the process of collecting stereology data will accelerate scientific research, toxicology studies and drug development, and bring significant benefits to society through scientific breakthroughs and medical discoveries. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

NSF Program Director: Erik Pierstorff