|
Research on the application of multi-parametric MRI radiomics based on different regions of interest in the evaluation of benign and malignant prostate lesions
|
W. Zhao   , L. Zhang , L. Wang , L. Chen , D. Li |
| Department of Radiology, Huaibei People's Hospital, Huaibei, Anhui, China , 452473418@qq.com |
|
|
Abstract: (20 Views) |
Background: The aim was to examine the diagnostic utility of radiomics via multi-parametric magnetic resonance imaging (mpMRI) in distinguishing between benign and malignant prostate lesions when features are extracted from different regions of interest (ROIs). Materials and Methods: A retrospective cohort of 99 individuals having mpMRI (T2WI, DCE, ADC) with histopathological confirmation was analyzed and split into training (n = 69) and validation (n = 30) at random as 7:3. Radiomics characteristics were taken from two ROIs: the lesion area (Set 1) and the whole-prostate area (Set 2). After choice of features by least absolute shrinkage and selection operator (LASSO), logistic regression (LR) was utilized to generate three models: Model 1 (lesion-based radiomics), Model 2 (prostate-based radiomics), and Model 3 (combined radiomics plus clinical factors). Receiver operating characteristic (ROC) was utilized to evaluate diagnostic results, and inter-model comparisons were conducted with DeLong’s test. Results: Of the 99 patients, 59 had benign and 40 had malignant lesions. In the training set, the area under the curve (AUC) values of Models 1, 2, and 3 were 0.997 (0.991–1.000), 0.969 (0.937–1.000), and 0.999 (0.997–1.000), while the corresponding validation AUCs were 0.920 (0.829–1.000), 0.916 (0.820–1.000), and 0.929 (0.843–1.000). There were no statistically significant variations among the three models. Conclusion: Radiomics models derived from mpMRI, regardless of whether features are extracted from the lesion area or from the entire prostate, demonstrate strong diagnostic capability in differentiating between prostate tumors that are benign and cancerous, suggesting substantial potential for clinical application. |
|
| Keywords: Prostate cancer, radiomics, machine learning, multiparametric magnetic resonance imaging. |
|
|
Full-Text [PDF 1443 kb]
(8 Downloads)
|
|
Type of Study: Original Research |
Subject:
Radiation Biology
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
