What question did this study set out to answer?

The study aims to predict tumor response to neoadjuvant chemoradiotherapy in rectal cancer using a machine learning-based radiomic approach.

May 22, 2026Open Access

MRI Based Radiomics as an Imaging Biomarker for Locally Advanced Carcinoma Rectum: Predicting Tumor Response Following Neoadjuvant Chemoradiotherapy as an Organ Preservation Strategy

Key Points

The study aims to predict tumor response to neoadjuvant chemoradiotherapy in rectal cancer using a machine learning-based radiomic approach.
Analyzed 100 patients with locally advanced rectal cancer undergoing pre- and post-treatment MRI after neoadjuvant chemoradiotherapy.
Extracted 124 radiomic features using TexRad software from T2W MRI and selected features with the recursive feature elimination algorithm.
Developed a prediction model using five machine learning classifiers, including random forest, with a focus on performance evaluation.
Selected five radiomic features for predicting tumor response using the random forest algorithm.
The random forest model achieved an AUC of 0.79 ± 0.15, accuracy of 0.72 ± 0.12, and precision of 0.77 ± 0.10 for the test cohort.
The radiomic model was found to be a non-invasive method for predicting tumor response in carcinoma rectum patients undergoing treatment.

Abstract

BACKGROUND: Patients with locally advanced rectal cancer show variable responses to neoadjuvant chemoradiotherapy (NACTRT), making early prediction of treatment response clinically important. Radiomics and machine learning may offer a potential non-invasive imaging biomarker to identify patients who are more likely to respond to treatment. AIM: We investigated a machine learning-based radiomic approach to predict tumor response (TR) to NACTRT in patients with adenocarcinoma of the rectum from baseline and post-NACTRT magnetic resonance imaging (MRI). METHODS: One hundred patients with locally advanced rectal cancer, who underwent pre- and post-treatment MRI after NACTRT were included in the study after approval from the institutional ethics committee. One hundred twenty-four radiomic features were extracted using the TexRad softwareTm, from pre- and post-treatment T2W MRI, which included first-order, GLCM, and shape features. Features were selected by the recursive feature elimination (RFE) algorithm. Five machine learning classifiers-support vector machine (SVM), random forest (RF), gradient boost (GB), Naive Bayes (NB) and AdaBoost (AB) classifiers were used to develop a prediction model for predicting TR. RESULTS: Five radiomic features were selected using for TR using RFE using RF algorithm. The RF model with 10-fold internal cross-validation performed the best for predicting TR. The predictive performance of the RF model for TR was found to be AUC: 0.79 ± 0.15, accuracy: 0.72 ± 0.12, and precision: 0.77 ± 0.10 for the test cohort respectively. CONCLUSION: RF classifier based radiomic model provided a non-invasive method to predict TR in patients of carcinoma rectum undergoing NACTRT treatment. The predictive potential of the radiomic features further requires validation in a larger cohort of patients in a multicentric study for clinical translation.

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper