The purpose of this work was to develop novel optical imaging technology and algorithms as a nondestructive method for detection and diagnosis of cancer in breast specimens. There are many ways in which the diagnosis of disease can benefit from fast and intelligent optical imaging technology. Our existing ability to provide this diagnosis depends on time-consuming pathology analysis. Optical coherence tomography (OCT) is a non-invasive optical imaging modality that provides depth-resolved, high-resolution images of tissue microstructure in real-time. OCT could provide a rapid evaluation of specimens while patients are still in the office, and has strong potential to improve the efficiency in evaluation of breast pathology specimens (biopsy or surgical).
In this work, we demonstrate an imaging system to address this unmet clinical need, artificial intelligence algorithms to interpret the images, and early work towards miniaturizing the technology.
We present an OCT system that achieves a line scan rate of 250kHz, meaning we can image a pathology cassette in 41 seconds, which is more than double the fastest scan rate in the field. By utilizing a multiplexed superluminescent diode (SLD) light source, which has strong noise performance over imaging speed, we achieve high resolution imaging under 5 um in tissue (axially and laterally). The system features a 1.1 mm 6-dB sensitivity fall-off range when imaging at 250 kHz. The scanner features large-area scanning with the implementation of a 2-axis motorized stage, enabling visualization of areas up to 10 cm x 10 cm (prior work visualizes 3 mm x 3mm). We showcase the results of demonstrating the performance of this system on a 100-patient clinical imaging study of breast biopsies, as well as imaging of clinical pathology specimens from the breast, prostate, lung, and pancreas in an IRB-approved study.
Further, we show our work towards developing artificial intelligence (AI) for cancer detection within OCT images. Using retrospective data, we developed a type of AI algorithm known as a convolutional neural network (CNN) to classify OCT images of breast tissue from 49 patients. The binary cancer classification achieved 94% accuracy, 96% sensitivity, and 92% specificity. This framework had higher accuracy than the 88% accuracy of 7 clinician readers combined in our lab’s earlier multi-reader study.
Lastly, we demonstrate a supercontinuum light source based on a 1 mm2 Si3N4 photonic chip for OCT imaging that has better performance than the state-of-the-art laser. Existing broadband laser sources for OCT are large, bulky, and have high excess noise. Our Si3N4 chip fundamentally eliminates the excess noise common to lasers and achieves 105 dB sensitivity and 1.81 mm 6-dB sensitivity roll-off with only 300 µW power on the sample.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/d8-6k52-4906 |
| Date | January 2021 |
| Creators | Mojahed, Diana |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0024 seconds