Automation of the ACR MRI Low-Contrast Resolution Test Using Machine Learning

Author(s)

Ramos, Jhonata E. ; Kim, Hae Yong ; Tancredi, F. B. ; Li, W ; Li, Q ; Wang, L

• Document type: Journal article
• Source: 2018 11TH INTERNATIONAL CONGRESS ON IMAGE AND SIGNAL PROCESSING, BIOMEDICAL ENGINEERING AND INFORMATICS (CISP-BMEI 2018); v. N/A, p. 6-pg., 2018-01-01.

Abstract

Magnetic Resonance Imaging (MRI) is a powerful, widespread and indispensable medical imaging modality. The American College of Radiology (ACR) recommends weekly acquisition of phantom images to assess the quality of scanner. Usually, these images must be analyzed by experienced technicians. Automatic analysis of these images would reduce costs and improve repeatability. Some automated methods have been proposed, but the automation of two of the ACR image quality tests remains open problem. Reports on the high-and low-contrast resolution tests are scarce and so far none of the proposed methods produce results robust enough to allow replacing human work. We use Machine Learning to emulate, with high accuracy, the detection of 120 low-contrast structures of ACR phantom by an experienced professional. We used a database with 620 sets of ACR phantom images that were acquired on scanners of different vendors, fields and coils, totaling 74,400 low-contrast structures. Technicians with more than 10 years of experience labeled each structure as 'detectable' or 'undetectable'. Machine learning algorithms were fed with image features extracted from the structures and their surroundings. Among the five methods we tested, Logistic Regression yielded the largest area under the ROC curve (0.878) and the highest Krippendorff's alpha (0.995). The results achieved in this study are substantially better than those previously reported in the literature. They are also better than the classifications made by junior technicians (with less than 5 years of experience). This indicate that the ACR MRI low-contrast resolution test may be automated using Machine Learning. (AU)

• FAPESP's process: 15/27022-0 - Quantitative MRI of cerebral perfusion at high spatial resolution.
• Grantee: Felipe Brunetto Tancredi
• Support Opportunities: Regular Research Grants