Using machine learning and optical coherence tomography to generate a non contact biopsy images
Abstract/Contents
- Abstract
- In vivo Biomedical imaging is an important part of diagnostic, therapeutic medical treatment and research. The field is constantly searching for viable clinical tools at the micro-scale (single cell resolution). Optical Coherence Tomography (OCT) enables real-time imaging of living tissues in 3D up to 1mm. In this work we develop tools and models to better quantify the OCT signal, we then apply those tools on two exciting applications in living mice brain imaging and early detection of skin cancer. We have developed a model to accurately quantify the signals produced by endogenous scattering in OCT imaging. This model predicts distinct concentration-dependent signal trends that arise from the underlying physics of OCT detection. The relation between OCT signal and particle concentration is approximately linear at low concentrations. However, at higher concentrations, interference effects cause signal to increase with a square root dependence on the number of particles within a voxel. Predictions were validated by measuring OCT signals from phantom models as well as living animals. More generally, the model described herein may inform the interpretation of detected signals in modalities that rely on coherence-based detection or are susceptible to interference effects. We then build a theoretical model for speckle noise, one of OCT's most prominent issues preventing us from experiencing the full potential of this technology. Our theoretical analysis using Monte Carlo simulations indicates that pure angular compounding (a common technology for reducing speckle) can improve the signal-to-noise ratio by no more than a factor of 1.3. We conclude that speckle reduction using angular compounding is equivalent to spatial averaging. We then shift gears and present how the above tools could be used to detect microscopic movement in living brain. We introduce a novel OCT-based neuroimaging setting and accompanied feature segmentation algorithm enabling rapid, accurate, and high-resolution imaging of 700 mu depth across the mouse cortex, in-vivo. We demonstrate 3D reconstruction of microarchitectural elements in a whole cortical column. Features as Central Nervous System cell bodies density, volume, and average distance to the nearest cell can be quantified to better study the brain by this highly transformative and versatile platform. We developed a laser-based barcoding technology that allows us to achieve precise registration of OCT images with H&E pathology images. This approach enabled us to assemble a library of > 1000 OCT - H&E images pairs from human skin samples. We then train a generative adversarial neural network on the image library and demonstrate the ability to virtually generate realistic H&E images directly from OCT images. The combination of the above technologies opens the door to a virtual biopsy allowing clinicians to examine H&E sections of a suspected area but bypassing the intensive and expensive biopsy.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2021; ©2021 |
| Publication date | 2021; 2021 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Winetraub, Yonatan |
|---|---|
| Degree supervisor | de la Zerda, Adam |
| Thesis advisor | de la Zerda, Adam |
| Thesis advisor | Chu, Steven |
| Thesis advisor | Levitt, Michael, 1947- |
| Degree committee member | Chu, Steven |
| Degree committee member | Levitt, Michael, 1947- |
| Associated with | Stanford University, Biophysics Program |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Yonatan Winetraub. |
|---|---|
| Note | Submitted to the Biophysics Program. |
| Thesis | Thesis Ph.D. Stanford University 2021. |
| Location | https://purl.stanford.edu/hx805fq4885 |
Access conditions
- Copyright
- © 2021 by Yonatan Winetraub
- License
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...