Clinical and radiomics prediction of complete response in rectal cancer pre-chemoradiotherapy. (July 2022)

• Record Type: Journal Article
• Authors: Mbanu, Peter; Saunders, Mark P.; Mistry, Hitesh; Mercer, Joe; Malcomson, Lee; Yousif, Saif; Price, Gareth; Kochhar, Rohit; Renehan, Andrew G.; van Herk, Marcel; Osorio, Eliana Vasquez
• Is Part Of: Physics and imaging in radiation oncology. Volume 23(2022)
• Page Start: 48
• Page End: 53
• Publication Date: 2022-07
• Subjects: Rectal cancer -- Chemoradiotherapy -- Magnetic resonance imaging -- Radiomics -- Treatment response; Radiotherapy -- Periodicals; Radiation dosimetry -- Periodicals; Cancer -- Imaging -- Periodicals; Oncology -- Periodicals; 615.842
• DOI: 10.1016/j.phro.2022.06.010
• Languages: English
• ISSNs: 2405-6316

Abstract

Highlights: Prediction of clinical complete response is essential for organ preservation management in rectal cancer. The addition of clinical variables can improve radiomics prediction of clinical complete response in rectal cancer. Clinical variables will need to be continuously updated and monitored over time to avoid calibration drift. Abstract: Background and purpose: Patients with rectal cancer could avoid major surgery if they achieve clinical complete response (cCR) post neoadjuvant treatment. Therefore, prediction of treatment outcomes before treatment has become necessary to select the best neo-adjuvant treatment option. This study investigates clinical and radiomics variables' ability to predict cCR in patients pre chemoradiotherapy. Materials and methods: Using the OnCoRe database, we recruited a matched cohort of 304 patients (152 with cCR; 152 without cCR) deriving training (N = 200) and validation (N = 104) sets. We collected pre-treatment MR (magnetic resonance) images, demographics and blood parameters (haemoglobin, neutrophil, lymphocyte, alkaline phosphate and albumin). We segmented the gross tumour volume on T2 Weighted MR Images and extracted 1430 stable radiomics features per patient. We used principal component analysis (PCA) and receiver operating characteristic area under the curve (ROC AUC) to reduce dimensionality and evaluate the models produced. Results: Using Logistic regression analysis, PCA-derived combined model (radiomics plus clinical variables) gave a ROC AUC of 0.76 (95% CI: 0.69–0.83) in the training set and 0.68 (95% CI 0.57–0.79) in the validation set. The clinical only model achieved an AUC of 0.73 (95% CI 0.66–0.80) and 0.62 (95% CI 0.51–0.74) in the training and validation set, respectively. The radiomics model had an AUC of 0.68 (95% CI 0.61–0.75) and 0.66 (95% CI 0.56–0.77) in the training and validation sets. Conclusion: The predictive characteristics of both clinical and radiomics variables for clinical complete response remain modest but radiomics predictability is improved with addition of clinical variables.