Automating Unbiased Stereology Cell Counting Using Deep Learning

image

Project Summary

Unbiased stereology is the field of study concerned with the accurate quantitative analysis of objects in three-dimensional (3D) space. The accuracy is achieved through the use of sampling methods and geometric probes specifically designed to avoid all known sources of non-random error (bias) hen making estimates of first order stereology parameters [Volume, Surface Area, Length, Number] and their variation. For example, making an unbiased estimate of the total number of objects (e.g., cells, nuclei) requires several steps: define a reference space; sample the reference space in a systematic-random manner; count the number of stained objects using a 3-D virtual probe (disector) using unbiased counting rules; and scale the resultant count to the entire reference space. Thin focal plane sampling using the disector probe avoids the Corpuscle problem introduced by sampling and counting 3D objects based on their appearance on 2D planes (i.e., profiles) through the reference space. With manual stereology to obtain ground truth counts, a trained technician slowly focuses from the top to the bottom of each disector stack while counting each object of interest as it comes into focus. The technicians repeat this thin focal plane scanning process for 200 disector stacks spaced in a systematic-random manner through the entire reference space. The manual Unbiased Stereology is tedious, time-consuming, and error-prone.
Deep learning application to Unbiased Stereology allivate the burdon of cell counting while providing automatic cell counting with high accuracy comparing to other approaches. In the applications of deep learning to unbiased stereology, stacks of images through each disector volume (disector stacks) are captured automatically using standard computer-assisted stereology equipment (i.e., microscope with motorized three-axis stage and digital imaging Stereologer, SRC Biosciences, Tampa, FL)). Each disector stack consists of multiple z-axis planes with one micron spacing to train Deep learning models to segment cells on microscopy images, followed by post-processing and Unbiased Stereology cell counting

Talk / Presentation

  • Feb 2020 - Dr. Dmitry Goldgof, AI+X Corporate Forum at University of South Florida "Deep Stereology for Automatic Quantification of Brain Cells"
  • Oct 2019 - Dr. Dmitry Goldgof, IEEE System Man and Cybernetics Conference SMC2019, Bari, Italy, on "Automatic Cell Counting using Active Deep Learning and Unbiased Stereology"
  • Dec 2018 - Saeed Alahmari, IEEE International Conference on Machine Learning and Applications ICMLA2018, Orlando, FL, USA, on "Iterative Deep Learning Based Unbiased Stereology With Human-in-the-Loop"

Research Team

Dmitry Goldgof, PhD, Distinguished Professor and Vice Chair, Department of Computer Science and Engineering, University of South Florida
Lawrence Hall, PhD, Distinguished Professor, Department of Computer Science and Engineering, University of South Florida
Peter R. Mouton, PhD, Professor and Director and Chief Scientific Officer, SRC Biosciences
Hady Phoulady, PhD, Assitant Professor, California State University, Sacramento, CA
Saeed Alahmari , PhD candidate, Department of Computer Science and Engineering, University of South Florida
Palak Dave, PhD student, Department of Computer Science and Engineering, University of South Florida
Hunter Morera, PhD Student, Department of Computer Science and Engineering, University of South Florida
Raj Patel

Publications

Our Sponsors

image image image image