Author ORCID Identifier

0000-0002-1896-1446

Date of Graduation

8-2021

Document Type

Dissertation (PhD)

Program Affiliation

Medical Physics

Degree Name

Doctor of Philosophy (PhD)

Advisor/Committee Chair

Jingfei Ma, Ph.D.

Committee Member

Gaiane M. Rauch, MD Ph.D.

Committee Member

Mark D. Pagel, Ph.D.

Committee Member

Ken-Pin Hwang, Ph.D.

Committee Member

R. Jason Stafford, Ph.D.

Committee Member

Steven H. Lin, M.D. Ph.D.

Abstract

Triple Negative Breast Cancer (TNBC) is an aggressive subtype of breast cancer which lacks upregulated hormone receptors. Because of this, it is not vulnerable to clinically available targeted therapies. When treated with standard of care neoadjuvant systemic therapy (NAST), TNBC only shows approximately a 40% rate of pathologic complete response (pCR). A biomarker which could predict TNBC response to NAST early during treatment would be useful, as it would allow for non-responders to be triaged to alternative therapies and potentially allow for the treatment of responders to be de-escalated.

Quantitative Magnetic Resonance Imaging (MRI) may be used to probe and measure aspects of the perfusion, diffusion, and mechanical properties of a cancer and its surroundings. In the research setting, several quantitative MRI biomarkers have shown potential for early prediction of response in breast cancer. However, TNBC shows a unique image phenotype on both conventional MRI and MRI biomarkers of response. Several MRI biomarkers of response which show promise in other breast cancer subtypes are not useful for predicting response in TNBC. This, in combination with the clinical needs of TNBC, warrants the development of MRI biomarkers of response that are specific to TNBC. This rational supports a large, ongoing prospective trial of TNBC patients at our institution who underwent longitudinal multiparametric MRI at pretreatment, after 2 cycles of NAST and after 4 cycles of NAST. In this dissertation, MRI biomarkers from diffusion MRI, dynamic contrast-enhanced (DCE) MRI, and magnetic resonance elastography (MRE) were developed and applied as predictors of NAST response in the prospective trial cohort.

First, aspects of the tumor necrosis on pretreatment diffusion MRI and DCE MRI were investigated as potential predictors of response. Our study established that no associations were present between tumor necrosis and the treatment response in our study population, thus served as a caution in the field for physicians considering necrosis on MRI as a possible negative predictive biomarker.

Second, functional tumor volume (FTV), an existing biomarker of response in breast cancer based on DCE MRI contrast thresholds, was optimized for early prediction of NAST response in TNBC. Fast DCE MRI from pretreatment and cycle 4 MRI scans was leveraged to find an optimal contrast timing to improve the predictive performance of FTV. FTV contrast thresholds optimized over the TNBC cohort paralleled TNBC subtype analysis presented by other groups in previous reports. This external validation further supports the use of a TNBC-specific FTV tuning for prediction of NAST response.

Third, diffusion MRI measurements in the peritumoral region were developed and applied as predictors of NAST response. We found that maximum diffusion and the standard deviation of diffusion in peritumoral regions including fatty tissues were useful for prediction of NAST response.

Finally, a convolutional neural network (CNN)-based MRE inversion algorithm was developed for improved spatial resolution of breast cancer MRE. Because acquisition of ground truth MRE data is impossible, simulating MRE data via finite volume methods (FVM) was substituted in CNN training. The CNN-based inversion algorithm was validated through gel phantom measurements. Validation on in vivo breast MRE was performed by comparing stiffness measurements from different breast tissues between the CNN-based algorithm and the existing vendor algorithm. Both algorithms were able to effectively distinguish between the tumor and other breast tissues, though only the vendor algorithm was able to distinguish between fatty tissue and fibroglandular tissue.

In conclusion, quantitative MRI biomarkers of breast cancer were developed and show promise for early prediction of NAST response in TNBC. MRI biomarkers of necrosis were not seen to be useful, while TNBC-tuned FTV and diffusion MRI of the peritumoral region showed promise for this purpose. A CNN-based inversion algorithm shows potential for MRE with improved spatial resolution, though additional development is required.

Keywords

Treatment Response, Quantitative MRI, Machine Learning

Download

Share

COinS