Coregistration of serial breast MR studies for augmenting neoadjuvant chemotherapy response evaluation in breast cancer patients

Neoadjuvant chemotherapy (NAC) is gaining popularity as an important part of breast conservation surgery in breast cancer patients and MRI is the primary modality of choice for predicting residual disease after therapy. By comparing residual tumor sizes in the follow-up (F/U) MR scans with the pre-treatment tumor size from baseline (B/L) scan, weighted decisions can be made regarding appropriate changes to the chemo regimen or effectiveness of the treatment for the particular patient. But due to the deformable nature of the breast, its shape in breast MR acquisitions of the same patient in different studies (taken 2-4 weeks from baseline study) is significantly different. Thus, the exact correspondence between the B/L and F/U scans is lost. In the presence of multi-centric or multi-focal breast cancer, an objective evaluation of tumor regression may be challenging. Also, inflammation due to treatment, i.e. mastitis, is a common occurrence and may be misdiagnosed as residual disease. Transporting the clinical results of neoadjuvant chemotherapy to the operating table is also arduous. Clinical evaluation using MR is done when the patient is in the prone position, while in the operating room, the patient is lying on her back, in the supine position. The obvious difference in both coordinate frames combined with the elastic nature of the breast makes the transition a challenging problem.; In this dissertation, we proposed a registration framework to augment quantitative assessment of patient response to NAC. By estimating the motion between consecutive temporal volumes of the patient, the precise location of the B/L tumor in the F/U scan can be determined. For this, we use image-based volumetric approaches that address boundary mapping as well as internal mapping. An important criteria involving tumor volume preservation was also incorporated to avoid tumor scaling due to non-rigid transformations which may convey a false sense of positive or negative response to chemotherapy. Another framework for registration of prone and supine datasets is also outlined. The registration framework has been successfully tested against well-calibrated phantoms and synthetic images and diverse patient data reflecting typical clinical scenarios.