Identifying high-risk colon cancer on CT an a radiomics signature improve radiologist’s performance for T staging?
Eun Kyoung Hong, Zuhir Bodalal, Federica Landolf, Nino Bogveradze, Paula Bos, Sae Jin Park, Jeong Min Lee, Regina Beets-Tan
Abdominal Radiology (2022) 47:2739–2746.
Colon cancer is the third most common malignancy and the fourth most common cause of cancer-related death [1]. Accurate staging and identification of high-risk colon cancer is crucial to avoid overtreatment and unnecessary exposure of low-risk patients to higher morbidity of surgical resection after preoperative treatment [2]. Based on published works, radiomics can provide us with quantitative imaging information to describe the comprehensive morphological features of the tumor, which can help and guide clinicians [1, 2, 3, 4]. Regarding the role of radiomics in colon cancer, most studies have investigated the role in terms of survival, lymph node metastasis, and molecular/genetic subtype prediction. Not much is known about radiomics application in risk stratification for colon cancer [2, 4, 5].
The aim of this retrospective study [2] was, firstly, to assess the additional value of radiomics features that were automatically extracted from pre-operative colon cancer CT in distinguishing high-risk colon cancer patients. Secondly, the purpose was to develop a combined model integrating both radiomic and conventional CT staging features for the prediction of high-risk colon cancer.
The authors used a sample of 292 patients (205 from Seoul National University Hospital and 87 from Netherlands Cancer Institute) with colon cancer (85 patients with ≤ T2 and 207 patients with ≥ T3 primary tumor) who received surgical resection and with pre-operative imaging within 60 days of surgery. Exclusion criteria were: neoadjuvant therapy, inadequate pathologic data, multiple primary tumors, non-visible tumor on CT, and poor quality CT. The radiologists noted the presence of the tumor beyond the muscularis propria layer of tumors (cT3-T4), blinded by all the clinical and histological information. Further, 6048 radiomic features were automatically extracted, and after applying an unsupervised dimensionality reduction technique, orthogonal principal feature analysis, only 29 features were selected and included for further analysis.
Univariate and multivariate logistic regression analysis was performed to determine the significant association between imaging features, CT staging and pathologic staging of colon cancer in differentiating tumor staging ≤ T2 vs. ≥ T3. Afterwards, using logistic regression, elastic net regularization and 10-fold cross validation, the authors developed different predictive models for each outcome. ROC curve analysis was conducted and area under the curve (AUC) with DeLong confidence interval (CI) was calculated to evaluate the performance of each model.
Results show that radiologist’s performance in CT staging of colon cancer was significantly associated with high pT tumors from both univariate (OR = 6.16, p < 0.001) and multivariate (OR = 1.22, p < 0.001) logistic regression analysis. The AUC value of CT staging was 0.697 (95% CI 0.638-0.756). Moreover, when radiomic features were added to the model, AUC increased to 0.779 (95% CI 0.720-0.839). From the 10-fold cross-validation, CT staging demonstrated AUC of 0.628 (95% CI 0.558-0.689) and increased to 0.727 (95% CI 0.621-0.777) when radiomic features were included.
These results of this study show that CT radiomic features could improve the radiologist’s performance in T staging colon cancer on CT, and using the combined model using both CT staging and radiomic features demonstrated significantly better diagnostic performance [2]. But, the integration of these combined models and radiomic scores in a clinical validation stage would require extensive multicenter prospective studies to prove their generalizability.
Neoadjuvant chemotherapy for locally advanced colon cancer remains an evolving treatment paradigm [6]. The FOxTROT trial, an international randomized controlled trial of 1052 patients evaluating neoadjuvant chemotherapy for colon cancer, used CT as a reliable tool for stratifying high-risk patients. The trial reported that 98% of patients undergoing neoadjuvant chemotherapy underwent surgery compared with 99% of patients who did not have neoadjuvant treatment, a reassuringly similar rate [7].
CT is already a routine part of preoperative staging of colon cancer, but in the context of documenting extracolonic disease [6]. A meta-analysis reported a sensitivity of 90%, and specificity of 69%, in differentiating stage T3 and T4 tumors on CT [8], which currently serves as the major selection criteria for patients with high-risk colon tumors who will receive neoadjuvant chemotherapy [9].
In conclusion, accurate pre-surgical staging of colon cancer is crucial for pre-operative chemotherapy, but the addition of radiomics/artificial intelligence approaches can help to build models that hold significant promise for integrated diagnosis and clinical decision making.
References:
- Mármol I, Sánchez-de-Diego C, Pradilla Dieste A, Cerrada E, Rodriguez Yoldi MJ. Colorectal Carcinoma: A General Overview and Future Perspectives in Colorectal Cancer. Int J Mol Sci. 2017 Jan 19;18(1):197. doi: 10.3390/ijms18010197.
- Hong EK, Bodalal Z, Landolfi F, Bogveradze N, Bos P, Park SJ, Lee JM, Beets-Tan R. Identifying high-risk colon cancer on CT and a radiomics signature improve radiologist's performance for T staging? Abdom Radiol (NY). 2022 Aug;47(8):2739-2746. doi: 10.1007/s00261-022-03534-0.
- Moldovanu CG, Boca B, Lebovici A, Tamas-Szora A, Feier DS, Crisan N, Andras I, Buruian MM. Preoperative Predicting the WHO/ISUP Nuclear Grade of Clear Cell Renal Cell Carcinoma by Computed Tomography-Based Radiomics Features. J Pers Med. 2020 Dec 23;11(1):8. doi: 10.3390/jpm11010008.
- Yao X, Sun C, Xiong F, Zhang X, Cheng J, Wang C, Ye Y, Hong N, Wang L, Liu Z, Meng X, Wang Y, Tian J (2020) Radiomic signature-based nomogram to predict disease-free survival in stage II and III colon cancer. Eur J Radiol 131:109205. doi:10.1016/j.ejrad.2020.109205.
- Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, Ma ZL, Liu ZY (2016) Development and Validation of a Radiomics Nomogram for Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer. J Clin Oncol 34 (18):2157-2164. doi:10.1200/JCO.2015.65.9128.
- Body A, Prenen H, Latham S, Lam M, Tipping-Smith S, Raghunath A, Segelov E. The Role of Neoadjuvant Chemotherapy in Locally Advanced Colon Cancer. Cancer Manag Res. 2021 Mar 17;13:2567-2579. doi: 10.2147/CMAR.S262870.
- Seymour M.T., Morton D., on behalf of the International FOxTROT Trial Investigators FOxTROT: An international randomised controlled trial in 1052 patients (pts) evaluating neoadjuvant chemotherapy (NAC) for colon cancer. J. Clin. Oncol. 2019;37:3504. doi: 10.1200/JCO.2019.37.15_suppl.3504.
- Nerad E, Lahaye MJ, Maas M, Nelemans P, Bakers FC, Beets GL, Beets-Tan RG (2016) Diagnostic Accuracy of CT for Local Stag-ing of Colon Cancer: A Systematic Review and Meta-Analysis. AJR Am J Roentgenol 207 (5):984-995. doi: 10.2214/AJR.15.15785.
- Foxtrot Collaborative G (2012) Feasibility of preoperativechemotherapy for locally advanced, operable colon cancer: the pilot phase of a randomised controlled trial. Lancet Oncol 13(11):1152-1160. doi: 10.1016/S1470-2045(12)70348-0.
Moldovanu Claudia-Gabriela is a young radiologist who has just completed her residency program at Emergency County Clinical Hospital in Cluj-Napoca, Romania. Her main areas of interest are abdominal and urogenital radiology and the use of artificial intelligence in radiology. She has been actively involved in clinical imaging research, contributing as author/co-author to many projects leading to conference presentations, publications and awards. She also recently completed her Ph.D. thesis with research focused on the role of artificial intelligence in medical diagnosis.
Comments may be sent to: moldovanucg@gmail.com