Dimitra Pouli, MD1, Roger Jenkins, MD2, Travis B Sullivan, MS2, Kimberly M Rieger-Christ, PhD2, Irene M Georgakoudi, PhD1, Thomas Schnelldorfer, MD2. 1Tufts University, Department of Biomedical Engineering, 2Lahey Hospital & Medical Center
Objective of the technology.
For pancreatic cancer patients, treatment selection fundamentally relies on staging, with “under-staging” considered a common problem. Imaging modalities that can complement conventional white-light laparoscopy, are needed to detect more accurately small metastatic lesions. Two-photon microscopy (TPM) is a non-invasive imaging modality that is progressively incorporated in micro-endoscopes and can provide subcellular resolution imaging of dense, optically scattering tissues. Collagen fibers, the main constituents of tissue stroma, naturally produce a two-photon scattering signal, named Second Harmonic Generation (SHG). As cancer invasion affects the extracellular matrix architecture originally in the microscopic level, SHG measurements have the potential to better detect otherwise occult macroscopically cancer metastases. The goal of this project was to set the foundation for an in situ, real time diagnostic analysis technique that could provide accurate classification of healthy and diseased tissue SHG images.
Description of the technology and method of its use or application
Freshly excised biopsies from healthy parietal peritoneum and primary pancreatic neoplasms were acquired from 4 patients. The tissues were imaged employing a multiphoton laser scanning microscope to generate SHG images at 900nm excitation and 460±20 nm emission. 8 images per tissue type were evaluated utilizing image texture analysis techniques, namely gray level co-occurrence matrices. 4 textural features (Contrast, Correlation, Energy and Homogeneity) were computed to describe the uniformity, spatial distribution and occurrence of intensity variations within each image and were finally used for discriminant analysis.
Preliminary results.
Healthy tissues displayed higher Contrast and Correlation features, translating in repetitive, increased local intensity variations within the healthy images. These outcomes agree with the recurrent appearance of bright collagen fibers. On the other hand, diseased tissue images exhibited higher Homogeneity and Energy values, translating in a more uniform and overall unvaried intensity pattern, a finding matching the destruction of the healthy stroma within the tumorous tissues. The textural discriminant analysis resulted in 87.5 % accurate classification in distinguishing diseased from healthy tissues. Lastly the computational cost of the analytical approach was less than half a second to extract the desired textural information.
Conclusions / Future directions.
We have shown that textural analysis of SHG images is feasible in an ultra-fast manner and can help establish an accurate classification system to differentiate SHG images from human healthy peritoneal and pancreatic tumor tissues. As with further optimization methods the computational cost could be further decreased, our analytical approach is a viable candidate for real time diagnostics.