Quantitative analysis of high spectral and spatial resolution (HiSS) breast MRI /

Quantitative analysis of high spectral and spatial resolution (HiSS) breast MRI /

Saved in:

Bibliographic Details
Author / Creator:	Weiss, William, author.
Imprint:	2015. Ann Arbor : ProQuest Dissertations & Theses, 2015
Description:	1 electronic resource (133 pages)
Language:	English
Format:	E-Resource Dissertations
Local Note:	School code: 0330
URL for this record:	http://pi.lib.uchicago.edu/1001/cat/bib/10773096

Hidden Bibliographic Details
Other authors / contributors:	University of Chicago. degree granting institution.
ISBN:	9781321884630
Notes:	Advisors: Maryellen L. Giger; Gregory S. Karczmar Committee members: Hiroyuki Abe; Milica Medved. This item must not be sold to any third party vendors. Dissertation Abstracts International, Volume: 76-11(E), Section: B. English
Summary:	This dissertation investigates high spectral and spatial resolution magnetic resonance imaging (HiSS MRI) and its clinical use as an aid in breast cancer screening and cancer diagnosis. HiSS data includes proton-resonance spectra with details about the water and fat content not available in conventional MRI acquisitions. Each spectrum provides information about the local physiology of a single image voxel, and is therefore susceptible to variations when the tissue in that voxel undergoes a physiologic change (e.g. when becoming cancerous), as in the formation of microcalcifications or increase in deoxygenated blood. For example, the spectra from malignant tumors tend to contain distortions in the water resonance, manifesting as an overall broadened lineshape or entirely separate peaks. This work develops ways to identify and quantify these distortions and examines their use in distinguishing malignant from benign lesions. The relationship between water and fat content as a measurement of breast density is studied as well, motivated by breast density's role in cancer risk assessment. Two techniques were developed to quantify spectral distortions of the water resonance for application in distinguishing between malignant and benign lesions. The first approach identified the largest off-peak components in a lesion and studied the ability of off-peak component magnitudes to distinguish malignant from benign lesions. Off-peak components were identified as the largest component of the residual spectrum created by subtracting a fit Lorentzian lineshape from the spectrum associated with each voxel in a lesion. Due to tumor heterogeneity, a voxel-elimination scheme was implemented to discard low-magnitude off-peak components, increasing the signal-to-noise ratio of the analyzed data. With a patient database of 15 malignant lesions and 8 benign lesions, receiver operating characteristic (ROC) analysis performed on the average off-peak component magnitude in each lesion yielded an area under the ROC curve (AUC) of 0.83 (95% confidence intervals [0.61, 0.98]) indicating good distinction between malignant and benign lesions. The second approach to discrimination between benign and malignant lesions employed the full complex water spectra to create a metric for quantifying distortions in the entire spectrum based on dispersion vs. absorption (DISPA) analysis. The metric, which I refer to as total radial difference (TRD), is a summation of the deviations of the spectrum's DISPA plot from that of an ideal complex Lorentzian, and includes more detail about the entire spectrum than just a single off-peak component. ROC analysis performed on the average TRD in each lesion with the same database as the previous study yielded an AUC of 0.90 (95% confidence intervals [0.27, 0.83]), indicating potentially better performance than off-peak magnitude in distinguishing malignant from benign lesions. A technique for measuring breast density based on water-peak and fat-peak integrals was developed. After applying multiple image processing steps to isolate the breast tissue from the background, skin, and chest wall, the total amount of water and total amount of fat in the breasts were calculated as the sum of the respective integrals over all three-dimensional breast images. The breast density estimation is equal to the ratio of total water to total water and fat. The HiSS-based densities of 22 high-risk patients were compared to corresponding mammographic percent densities and radiologist-assigned BI-RADS density scores. HiSS density showed very low inter-user and intra-user variability and correlated well with BI-RADS density scores (p < 0.01), while no significant relationship was found between mammographic density and HiSS density (p > 0.05) or mammographic density and BI-RADS scores (p > 0.05). Our results demonstrate that data from high spectral and spatial resolution MRI may benefit multiple breast interpretation tasks. With further work to optimize HiSS image acquisition and reconstruction process as well as larger-scale clinical studies similar to those presented in this dissertation, HiSS MRI may become a viable supplement to current imaging methods.