Publication: Image quality evaluation of ACR (RMI 156) phantom using machine learning in digital breast tomosynthesis (DBT)
Loading...
Date
2025-06
Authors
Othman, Norsyafiqah Andriana
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Quality assurance (QA) typically relies on the RMI 156 Phantom to evaluate parameters like resolution, contrast, and noise, but current manual assessments are subjective, vary among evaluators, and are time-consuming, leading to inconsistencies. These limitations compromise the reliability of QA processes, potentially affecting diagnostic accuracy. This study explores the application of ML in automating image quality assessment for DBT using the ACR (RMI 156) phantom. The main objective was to develop an ML-based framework capable of evaluating image quality with improved accuracy, consistency, and efficiency compared to conventional manual methods. DBT images acquired from phantom exposures were processed using MATLAB, including preprocessing, segmentation, feature extraction, and data augmentation. Three classification models which are Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF) were trained and evaluated using 10-fold cross-validation. Results showed that all models achieved high accuracy, with RF slightly outperforming others. SVM demonstrated superior recall and F1 score, particularly in detecting minority class instances. Both KNN (0.10 Precision, 0.10 Recall, 0.1 F1 Score) and Rf (0 Precision, Recall, 0 F1 Score) achieved the high accuracy of 93.89%, followed by SVM (0.033 Precision, 0.10 Recall, 0.05 F1 Score), achieved the accuracy of 87.04%. In terms of training time, SVM (0.0149s) and KNN (0.0216s) were faster, while the RF model required more time (0.9198s) due to its ensemble structure. Despite achieving promising results, the study faced limitations such as dataset imbalance and the exclusion of clinical data. The findings suggest that ML offers a solution for DBT image quality control and recommend further research incorporating larger datasets and deep learning techniques to enhance generalisability and real-world applicability
Description
Keywords
-