|
Prediction of treatment response in inoperable cervical cancer treated with curative radiotherapy using artificial intelligence with radiomic and clinical parameters
|
M. Yakar   , D. Etiz , K. Duruer , E. Gulbandilar , E. Colak , B. Barut |
| Department of Radiation Oncology, Faculty of Medicine, Osmangazi University, Eskişehir, Turkey , myakar@ogu.edu.tr |
|
|
Abstract: (11 Views) |
Background: Radiomic features extracted from the tumor and peritumoral region can be used to predict response to radiotherapy (RT). This research sought to estimate treatment response through the analysis of tumor and peritumoral radiomic features in patients receiving curative radiotherapy for cervical cancer. Materials and Methods: Between 2015 and 2024, 60 patients with cervical cancer who underwent radiotherapy (RT) with or without chemotherapy (ChT) were included in the study Tumor segmentation was performed using computed tomography (CT) scans. Radiomic feature extraction was carried out with the LifeX platform (v7.0.0). A total of 110 tumoral and peritumoral radiomic features were obtained. 19 clinical and 110 radiomic features were evaluated, and important variables were identified using a decision tree algorithm. Treatment response following RT was evaluated based on the RECIST guidelines. The k-Nearest Neighbors (kNN) and Multi-Layer Perceptron (MLP) algorithms were used to build a prediction model. Results: The analysis identified total EQD2, EBRT interruption time, peritumoral_GLZLM_ZLNU, GTV_GLZLM_ZLNU, and GTV_GLRLM_LRLGE as significant predictive variables. Among the predictive algorithms developed for key variables, the k-Nearest Neighbors (kNN) model achieved F1 scores of 0.84 and 0.74 on the training and testing datasets, respectively, while the Multi-Layer Perceptron (MLP) model achieved F1 scores of 0.96 and 0.81 on the training and testing datasets, respectively. Conclusion: Non-invasive, tumor-signature radiomics show promise for guiding personalized therapies. The radiomic parameters Peritumoral_GLZLM_ZLNU, GTV_GLZLM_ZLNU, and GTV_GLRLM_LRLGE emerged as key predictors of treatment response to radiotherapy. Nevertheless, further multicenter research involving larger patient cohorts is required to achieve more robust and generalizable results. |
|
| Keywords: Uterine cervical neoplasms, radiomics, radiotherapy, chemoradiotherapy, prognosis, image processing - computer assisted. |
|
|
Full-Text [PDF 1220 kb]
(4 Downloads)
|
|
Type of Study: Original Research |
Subject:
Radiation Biology
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
