Anzeige
Erschienen in:
15.04.2020 | Hollow Organ GI
Clinical utility of radiomics at baseline rectal MRI to predict complete response of rectal cancer after chemoradiation therapy
Erschienen in: Abdominal Radiology | Ausgabe 11/2020
Einloggen, um Zugang zu erhaltenAbstract
Purpose
To investigate the value of T2-radiomics combined with anatomical MRI staging criteria from pre-treatment rectal MRI in predicting complete response to neoadjuvant chemoradiation therapy (CRT).
Methods
This retrospective study included patients with locally advanced rectal cancer who underwent rectal MRI before neoadjuvant CRT from October 2011 to January 2015 and then surgery. Surgical histopathologic analysis was used as the reference standard for pathologic complete response. Anatomical MRI staging criteria were extracted from our institutional standardized radiology report. In radiomics analysis, one radiologist manually segmented the primary tumor on T2-weighted images for all 102 patients (i.e., training set); two different radiologists independently segmented 66/102 patients (i.e., validation set). 108 radiomics features were extracted. Then, scanner-independent features were identified and least absolute shrinkage operator analysis was used to extract a radiomics score. Finally, a support vector machine model combining the radiomics score and anatomical MRI staging criteria was compared against both anatomical MRI-only and radiomics-only models using the deLong test.
Results
The study included 102 patients (42 women; median age = 61 years).The radiomics score produced an area under the curve (AUC) of 0.75. Comparable results were found using the validation set (AUCs = 0.75 and 0.71 for each radiologist, respectively). The anatomical MRI-only model had an accuracy of 67% (sensitivity 42%, specificity 72%); when adding the radiomics score, the accuracy increased to 74% (sensitivity 58%, specificity 77%).
Conclusion
Combining T2-radiomics and anatomical MRI staging criteria from pre-treatment rectal MRI may help to stratify patients based on the prediction of treatment response to neoadjuvant therapy.