Original Article
Integrated radiomic model for predicting the prognosis of esophageal squamous cell carcinoma patients undergoing neoadjuvant chemoradiation
Abstract
Background: To establish a feasible prediction model for prognoses of esophageal squamous cell carcinoma (ESCC) patients undergoing neoadjuvant concomitant chemoradiation (NACCRT).
Methods: Post-chemoradiation computed tomography (CT) radiomics features and clinical parameters were investigated. CT images from advanced thoracic ESCC patients treated with NACCRT and esophagectomy were extracted for radiomics features. Least absolute shrinkage and selection operator regression were used to select features and build signatures. Radiomics signatures and clinical factors were integrated into Cox regression analysis for prognosis; the prediction model’s performance was examined via receiver-operating characteristic (ROC) curve analysis.
Results: A total of 46 radiomics features and 25 clinical parameters were extracted from 62 cases, of which 59 passed image processing and became eligible for model testing. Eight selected radiomics features showed good prediction power [area under the curve (AUC) =0.851] and reliability in predicting pathological complete response (pCR). The radiomics signature and clinical parameter combination model showed increased prediction power of radiomics signature alone for local regional failure (LRF) (AUC=0.804) and distant failure (DF) (AUC=0.754). Following were the strongest contributors of prediction power for prognostic endpoints: (I) resection status multiplied by long-run emphasis in grey-level run length matrix (GLRLM_LRE) for progression (hazard ratio=8.776); (II) non-uniformity of the grey-levels (GLRLM_GLNU) (hazard ratio=6.888); and (III) sphericity (hazard ratio=0.152) for overall survival (OS).
Conclusions: The integrated prediction model for prognosis may aid clinicians in decision making regarding post-operative adjuvant therapy for ESCC patients undergoing NACCRT.
Methods: Post-chemoradiation computed tomography (CT) radiomics features and clinical parameters were investigated. CT images from advanced thoracic ESCC patients treated with NACCRT and esophagectomy were extracted for radiomics features. Least absolute shrinkage and selection operator regression were used to select features and build signatures. Radiomics signatures and clinical factors were integrated into Cox regression analysis for prognosis; the prediction model’s performance was examined via receiver-operating characteristic (ROC) curve analysis.
Results: A total of 46 radiomics features and 25 clinical parameters were extracted from 62 cases, of which 59 passed image processing and became eligible for model testing. Eight selected radiomics features showed good prediction power [area under the curve (AUC) =0.851] and reliability in predicting pathological complete response (pCR). The radiomics signature and clinical parameter combination model showed increased prediction power of radiomics signature alone for local regional failure (LRF) (AUC=0.804) and distant failure (DF) (AUC=0.754). Following were the strongest contributors of prediction power for prognostic endpoints: (I) resection status multiplied by long-run emphasis in grey-level run length matrix (GLRLM_LRE) for progression (hazard ratio=8.776); (II) non-uniformity of the grey-levels (GLRLM_GLNU) (hazard ratio=6.888); and (III) sphericity (hazard ratio=0.152) for overall survival (OS).
Conclusions: The integrated prediction model for prognosis may aid clinicians in decision making regarding post-operative adjuvant therapy for ESCC patients undergoing NACCRT.
