A radiomics model based on transrectal ultrasound for predicting prostate cancer
DOI:
https://doi.org/10.11152/mu-4329Keywords:
prostate cancer, transrectal ultrasound, radiomics, machine learning, image segmentationAbstract
Aim: Prostate cancer (PCa) is one of the most common neoplasms in men. However, the value of ultrasound-based radiomics for diagnosing PCa remains uncertain.
Material and methods: We retrospectively analyzed ultrasonic and clinical data from 373 patients. Patients were divided into two groups according to the pathological results. Radiomics features were
extracted from TRUS, and we screened the optimal features to construct radiomics models. Relationships between clinical characteristics and prostate lesions were identified by univariate and multivariate logistic regression analysis. Finally, a clinical-radiomics model was developed, and then visualized in the form of a nomogram.
Results: Of the 373 patients, 178 had benign disease and 195 had malignant disease. The support vector machine (SVM) classification model showed the best performance, while the diagnostic performance of the clinical model was poorer than that of the radiomics model (p<0.05) or the combined (clinical-radiomics) model (p<0.05). In general, the combined model demonstrated the highest AUC and proved to be more advantageous.
Conclusion: The prediction model we constructed based on TRUS predicted PCa preoperatively with high efficiency. In addition, combining radiomics with clinical factors improved diagnostic accuracy.
Downloads
Published
Issue
Section
License
All rights are reserved to the Medical Publishing House of the „Iuliu Hatieganu" University of Medicine and Pharmacy, Cluj-Napoca; no parts of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without the permission of the editor.
Correspondence regarding permission to reprint all or part of any article published in this journal should be addressed to the Editor, e-mail: medultrasonography@gmail.com.


