Premium
SU‐F‐R‐16: Delineation of Tumor Habitats Based On Dynamic Contrast Enhanced MRI
Author(s) -
Chang Y,
Solorzano G,
Ackerstaff E,
Yechieli R,
Subhawong T,
Stoyanova R
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4955788
Subject(s) - principal component analysis , voxel , contrast (vision) , computer science , pattern recognition (psychology) , artificial intelligence , noise (video) , nuclear medicine , medicine , image (mathematics)
Purpose: To determine an automatic procedure for determining the number and delineation of tumor habitats from Dynamic Contrast Enhanced (DCE) MRI data. Methods: An automatic procedure for delineation of tumor habitats from DCE‐MRI datasets was developed as a two‐part process involving: (1) statistical testing of the principal components of the dataset to detect the number of habitats and (2) the use of this number in an unsupervised pattern recognition (PR) technique to recover the temporal pattern and location of detected habitats. The procedure was tested on simulated DCE‐MRI datasets with one, two, and three unique temporal patterns, corresponding to well‐perfused, hypoxic, and necrotic tumor environments, at various noise levels. Recovery of the temporal pattern and location of detected habitats was tested on a separate simulated DCE‐MRI dataset mimicking tumors with three unique habitats present in mixed proportions in each voxel of the dataset at different noise levels. The automatic procedure for delineation of tumor habitats was then applied to a DCE‐MRI dataset from a preclinical prostate cancer tumor model and a soft tissue adult fibrosarcoma. Results: Statistical testing of the principal components of the simulated DCE‐MRI datasets revealed correct estimation of the number of habitats in 99.6%, 99.5%, 99.7%, and 99.6% of trials (1000 total trials) when signal‐to‐noise ratio was 10, 7.5, 5, and 2.5, respectively. Recovery of the temporal pattern and location of tumor habitats from simulated DCE‐MRI datasets using the unsupervised PR technique produced the three habitats used in the simulation with good fidelity at all noise levels. The automatic procedure for delineation of tumor habitats revealed three unique temporal patterns in the experimental rat model, resembling wellperfused, hypoxic, and necrotic habitats. Conclusion: The number of underlying tumor habitats can be determined using the automatic procedure, allowing potential use for finding hypoxic habitats indicative of worse patient outcomes.