Image registration, Cancer, Localisation, MR, Computer applications, Breast, Artificial Intelligence
L. Cherkezyan PhD1, E. Aribal2, O. Buğdaycı2, C. Caluser3, Y. Lei4, Z. Anwar5; 1Evanston/US, 2Istanbul/TR, 3Glen Ellyn, IL/US, 4Chicago/US, 5Napperville/US
Aims and objectives
Targeted second-look ultrasound scan is frequently prescribed for patients with MRI-detected small breast lesions to increase MRI specificity or enable ultrasound guided biopsies and other procedures.
Despite its clinical relevance,
mapping MRI-detected small lesions that are obtained with the patient in prone position to their probable location while the patient is supine during an ultrasound scan,
remains a significant challenge due to the deformation of the breast.
second-look ultrasound identifies 57.5% of MRI detected lesions at a pooled rate .
Many mapping methods have been proposed,
but extensive computational time and/or limited precision prevent their clinical application.
Modeling mechanical behavior of the breast using machine learning (ML) appears to be a feasible approach to solve the problem .
Our method proposes using ML to provide fast breast lesion localization in supine position for ultrasound scan,
from their location information obtained on prone MRI images,
to guide ultrasound localization or other procedures done in supine position.