Publication: Fabrication of angiography quality control phantom for image quality evaluation using machine learning
| dc.contributor.author | Aziz, Muhammad Haziq Abd | |
| dc.date.accessioned | 2025-10-16T06:36:58Z | |
| dc.date.available | 2025-10-16T06:36:58Z | |
| dc.date.issued | 2025-06 | |
| dc.description.abstract | Angiography's QC suffers from subjective evaluations and a lack of specialised phantoms. This study addresses this by developing an affordable, in-house angiography phantom and evaluating the image quality using a machine learning (ML) approach. Purpose: 1) Design and fabricate an in-house phantom for high contrast and spatial resolution; 2) Assess ML model performance and validation; 3) Validate the best ML for evaluation of phantom image quality. Method: An in-house phantom was 3D-printed using LW-PLA-HT, incorporating tungsten carbide beads for high contrast and a Huttner Type 18-line pair for spatial resolution. 14 angiographic images were acquired from HPUSM and analysed in MATLAB R2024a. Image analysis involved pre-processing, segmentation, feature extraction and augmentation were applied. Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF) classifiers were evaluated using accuracy, precision, recall, F1-score, and AUC, with 10-fold cross-validation and an 80/20 training/testing. Results: Human evaluations showed variability. Among SVM, KNN, and RF, Random Forest demonstrated the best overall performance. For high-contrast image classification, RF achieved exceptional results (100% accuracy, 1.0000 F1 score), followed by KNN (76.11% accuracy, 0.7503 F1 score), and SVM (61.95% accuracy, 0.6095 F1 score). Spatial resolution classification was more challenging, with RF again leading (90.32% accuracy, 0.9050 F1 score), followed by KNN (64.52% accuracy, 0.6650 F1 score), and SVM (32.26% accuracy, 0.3180 F1 score). Conclusion: Random Forest demonstrated the best performance in this research, which highlights the viability of fabricating a cost-effective angiography phantom and utilising ML for image quality assessment. | |
| dc.identifier.uri | https://erepo.usm.my/handle/123456789/22837 | |
| dc.language.iso | en | |
| dc.subject | - | |
| dc.title | Fabrication of angiography quality control phantom for image quality evaluation using machine learning | |
| dc.type | Resource Types::text::thesis::bachelor thesis | |
| dspace.entity.type | Publication | |
| oairecerif.author.affiliation | Universiti Sains Malaysia |