Radiomics analysis and supervised machine learning model for classification of cervical cancer images using diffusion weighted imaging-MRI

Cervical cancer is the third most prevalent cause of mortality among women in Malaysia. Early detection, especially in high-risk populations, can reduce mortality rates and enable timely treatment. This study investigates the efficacy of staging classification using diffusion-weighted imaging magnetic resonance imaging (DWIMRI) through radiomic analysis and machine learning. Data were retrospectively analyzed from the picture archiving and communication system (PACS) at Institut Kanser Negara (IKN) in Putrajaya, Malaysia. The first objective involved 30 patients to evaluate the repeatability and reproducibility of manual and semi-automated segmentation methods on DWI-MRI images. Intra-class correlation coefficient (ICC) analyses were performed on 662 radiomic features encompassing texture, shape, and first-order statistics. The semi-automated active contour model (ACM) algorithm (average ICC = 0.952 ± 0.009, p > 0.05) was found to be more robust and reproducible than fully manual segmentation (average ICC = 0.897 ± 0.011, p > 0.05). The second objective assessed the stability of radiomic features using contrast-limited adaptive histogram equalization (CLAHE) for image enhancement of 80 DWI-MRI images, enhanced images exhibited improved stability in radiomic features (ICC = 0.990 ± 0.005, p < 0.05), outperforming both semi-automated (ICC = 0.864 ± 0.033, p < 0.05) and manual methods (ICC = 0.554 ± 0.185, p > 0.05). The third objective focused on classifying cervical cancer stages using DWI-MRI radiomic features. A support vector machine (SVM) classifier yielded excellent performance metrics, accuracy of 0.77, and precision of 0.63, with an area under the curve (AUC) of 96%. Additionally, the SVM algorithm was evaluated based on its performance across different DWI bvalues, aiming to optimize scanning time. In conclusion, SVM-based models can develop accurate and reproducible software for classifying cervical cancer stages, significantly enhancing the role of radiology by enabling more quantitative MRI interpretations. This study underscores the potential of radiomic analysis to improve the accuracy of medical reports, reduce dependency on contrast agents, and enhance early detection of cervical cancer.

Text 119334.pdf Download (8MB) Official URL or Download Paper: http://ethesis.upm.edu.my/id/eprint/18442

Actions (login required)

View Item