Selecting Candidates for Organ‐Preserving Strategies After Neoadjuvant Chemoradiotherapy for Rectal Cancer: Development and Validation of a Model Integrating MRI Radiomics and Pathomics

作者：Lijuan Wan, Sun Zhuo, Wenjing Peng, Sicong Wang, Jiangtao Li, Qing Zhao, Shuhao Wang, Han Ouyang, Xinming Zhao, Shuangmei Zou, Hongmei Zhang · 发表于：Journal of Magnetic Resonance Imaging · 年份：2022 · DOI：10.1002/jmri.28108 · 被引用次数：39 · 研究领域：Colorectal Cancer Surgical Treatments、Radiomics and Machine Learning in Medical Imaging、MRI in cancer diagnosis

BACKGROUND: Histopathologic evaluation after surgery is the gold standard to evaluate treatment response to neoadjuvant chemoradiotherapy (nCRT) in locally advanced rectal cancer (LARC). However, it cannot be used to guide organ-preserving strategies due to poor timeliness. PURPOSE: To develop and validate a multiscale model incorporating radiomics and pathomics features for predicting pathological good response (pGR) of down-staging to stage ypT0-1N0 after nCRT. STUDY TYPE: Retrospective. POPULATION: A total of 153 patients (median age, 55 years; 109 men; 107 training group; 46 validation group) with clinicopathologically confirmed LARC. FIELD STRENGTH/SEQUENCE: -weighted and single-shot EPI diffusion-weighted images. ASSESSMENT: The differences in clinicoradiological variables between pGR and non-pGR groups were assessed. Pretreatment and posttreatment radiomics signatures, and pathomics signature were constructed. A multiscale pGR prediction model was established. The predictive performance of the model was evaluated and compared to that of the clinicoradiological model. STATISTICAL TESTS: test, Fisher's exact test, t-test, the minimum redundancy maximum relevance algorithm, the least absolute shrinkage and selection operator logistic regression algorithm, regression analysis, receiver operating characteristic curve (ROC) analysis, Delong method. P < 0.05 indicated a significant difference. RESULTS: Pretreatment radiomics signature (odds ratio [OR] = 2.53; 95% CI: 1.58-4.6...